0.0.149

* fix tool errors

*QA

.github/workflows/pullfrog.yml

@@ -29,7 +29,9 @@ jobs:
       # - uses: actions/setup-node@v6
 
       - name: Run agent
-        uses: pullfrog/action@v0
+        uses: pullfrog/action@main
+        env: 
+          log_level: ${{ vars.LOG_LEVEL }}
         with:
           prompt: ${{ github.event.inputs.prompt }}
 

.husky/pre-commit

@@ -1,12 +1,13 @@
-# Ensure lockfile is up to date if package.json changed
+# Check if lockfile needs updating
 if git diff --cached --name-only | grep -q "^package.json$"; then
     echo "🔒 Updating lockfile..."
     pnpm lock
+    git add pnpm-lock.yaml
+fi
 
-    # Build the action before committing
+# Check if entry needs rebuilding (entry.ts, esbuild.config.js, or any .ts files)
+if git diff --cached --name-only | grep -qE "^(entry\.ts|esbuild\.config\.js|.*\.ts)$"; then
     echo "🔨 Building action..."
     pnpm build
-
-    # Add the built files and lockfile to the commit
-    git add entry pnpm-lock.yaml
+    git add entry
 fi

agents/claude.ts

@@ -14,12 +14,12 @@ export const claude = agent({
       executablePath: "cli.js",
     });
   },
-  run: async ({ payload, mcpServers, apiKey, cliPath }) => {
+  run: async ({ payload, mcpServers, apiKey, cliPath, repo }) => {
     // Ensure API key is NOT in process.env - only pass via SDK's env option
     delete process.env.ANTHROPIC_API_KEY;
 
-    const prompt = addInstructions(payload);
-    console.log(prompt);
+    const prompt = addInstructions({ payload, repo });
+    log.group("» Full prompt", () => log.info(prompt));
 
     // configure sandbox mode if enabled
     const sandboxOptions: Options = payload.sandbox
@@ -63,6 +63,7 @@ export const claude = agent({
 
     // Stream the results
     for await (const message of queryInstance) {
+      log.debug(JSON.stringify(message, null, 2));
       const handler = messageHandlers[message.type];
       await handler(message as never);
     }

agents/codex.ts

@@ -20,7 +20,7 @@ export const codex = agent({
       executablePath: "bin/codex.js",
     });
   },
-  run: async ({ payload, mcpServers, apiKey, cliPath }) => {
+  run: async ({ payload, mcpServers, apiKey, cliPath, repo }) => {
     // create config directory for codex before setting HOME
     const tempHome = process.env.PULLFROG_TEMP_DIR!;
     const configDir = join(tempHome, ".config", "codex");
@@ -61,7 +61,7 @@ export const codex = agent({
     );
 
     try {
-      const streamedTurn = await thread.runStreamed(addInstructions(payload));
+      const streamedTurn = await thread.runStreamed(addInstructions({ payload, repo }));
 
       let finalOutput = "";
       for await (const event of streamedTurn.events) {

agents/cursor.ts

@@ -91,7 +91,7 @@ export const cursor = agent({
       executableName: "cursor-agent",
     });
   },
-  run: async ({ payload, apiKey, cliPath, mcpServers }) => {
+  run: async ({ payload, apiKey, cliPath, mcpServers, repo }) => {
     configureCursorMcpServers({ mcpServers, cliPath });
     configureCursorSandbox({ sandbox: payload.sandbox ?? false });
 
@@ -166,7 +166,8 @@ export const cursor = agent({
     };
 
     try {
-      const fullPrompt = addInstructions(payload);
+      const fullPrompt = addInstructions({ payload, repo });
+      log.group("» Full prompt", () => log.info(fullPrompt));
 
       // configure sandbox mode if enabled
       // in sandbox mode: remove --force flag and rely on cli-config.json sandbox settings
@@ -211,6 +212,7 @@ export const cursor = agent({
 
           try {
             const event = JSON.parse(text) as CursorEvent;
+            log.debug(JSON.stringify(event, null, 2));
 
             // skip empty thinking deltas
             if (event.type === "thinking" && event.subtype === "delta" && !event.text) {
@@ -307,7 +309,7 @@ function configureCursorMcpServers({ mcpServers }: ConfigureMcpServersParams) {
   }
 
   writeFileSync(mcpConfigPath, JSON.stringify({ mcpServers: cursorMcpServers }, null, 2), "utf-8");
-  log.info(`MCP config written to ${mcpConfigPath}`);
+  log.info(`» MCP config written to ${mcpConfigPath}`);
 }
 
 /**
@@ -346,5 +348,5 @@ function configureCursorSandbox({ sandbox }: { sandbox: boolean }): void {
       };
 
   writeFileSync(cliConfigPath, JSON.stringify(config, null, 2), "utf-8");
-  log.info(`CLI config written to ${cliConfigPath} (sandbox: ${sandbox})`);
+  log.info(`» CLI config written to ${cliConfigPath} (sandbox: ${sandbox})`);
 }

agents/gemini.ts

@@ -154,15 +154,15 @@ export const gemini = agent({
       ...(githubInstallationToken && { githubInstallationToken }),
     });
   },
-  run: async ({ payload, apiKey, mcpServers, cliPath }) => {
+  run: async ({ payload, apiKey, mcpServers, cliPath, repo }) => {
     configureGeminiMcpServers({ mcpServers, cliPath });
 
     if (!apiKey) {
       throw new Error("google_api_key or gemini_api_key is required for gemini agent");
     }
 
-    const sessionPrompt = addInstructions(payload);
-    log.info(`Starting Gemini CLI with prompt: ${payload.prompt.substring(0, 100)}...`);
+    const sessionPrompt = addInstructions({ payload, repo });
+    log.group("» Full prompt", () => log.info(sessionPrompt));
 
     // configure sandbox mode if enabled
     // --allowed-tools restricts which tools are available (removes others from registry entirely)
@@ -184,6 +184,7 @@ export const gemini = agent({
     }
 
     let finalOutput = "";
+    let stdoutBuffer = ""; // buffer for incomplete lines across chunks
     try {
       const result = await spawn({
         cmd: "node",
@@ -195,8 +196,13 @@ export const gemini = agent({
           const text = chunk.toString();
           finalOutput += text;
 
-          // parse each line as JSON (gemini cli outputs one JSON object per line)
-          const lines = text.split("\n");
+          // buffer incomplete lines across chunks (NDJSON format)
+          stdoutBuffer += text;
+          const lines = stdoutBuffer.split("\n");
+
+          // keep the last element (may be incomplete) in the buffer
+          stdoutBuffer = lines.pop() || "";
+
           for (const line of lines) {
             const trimmed = line.trim();
             if (!trimmed) continue;
@@ -210,8 +216,8 @@ export const gemini = agent({
                 await handler(event as never);
               }
             } catch {
-              console.log("parse error", trimmed);
               // ignore parse errors - might be non-JSON output from gemini cli
+              log.debug(`[gemini] non-JSON stdout line: ${trimmed.substring(0, 200)}`);
             }
           }
         },

agents/index.ts

@@ -6,6 +6,8 @@ import { gemini } from "./gemini.ts";
 import { opencode } from "./opencode.ts";
 import type { Agent } from "./shared.ts";
 
+export type { Agent } from "./shared.ts";
+
 export const agents = {
   claude,
   codex,

agents/instructions.ts

@@ -1,93 +1,61 @@
+import { execSync } from "node:child_process";
 import { encode as toonEncode } from "@toon-format/toon";
 import type { Payload } from "../external.ts";
 import { ghPullfrogMcpName } from "../external.ts";
 import { getModes } from "../modes.ts";
 
+interface RepoInfo {
+  owner: string;
+  name: string;
+  defaultBranch: string;
+}
+
 /**
- * Extract only essential fields from event data to reduce token usage.
- * Removes verbose GitHub API metadata (user objects, repository metadata, etc.)
- * and keeps only the fields agents actually need.
+ * Build runtime context string with git status, repo data, and GitHub Actions variables
  */
-// function extractEssentialEventData(event: Payload["event"]): Record<string, unknown> {
-//   const trigger = event.trigger;
-//   const essential: Record<string, unknown> = { trigger };
+function buildRuntimeContext(repo: RepoInfo): string {
+  const lines: string[] = [];
 
-//   // common fields
-//   if ("issue_number" in event) {
-//     essential.issue_number = event.issue_number;
-//   }
-//   if ("branch" in event && event.branch) {
-//     essential.branch = event.branch;
-//   }
+  // working directory
+  lines.push(`working_directory: ${process.cwd()}`);
+  lines.push(`log_level: ${process.env.LOG_LEVEL}`);
 
-//   // trigger-specific fields
-//   switch (trigger) {
-//     case "issue_comment_created":
-//       if ("comment_id" in event) essential.comment_id = event.comment_id;
-//       if ("comment_body" in event) essential.comment_body = event.comment_body;
-//       // include issue title/body if available in context (but not the entire context object)
-//       if ("context" in event && event.context && typeof event.context === "object") {
-//         const ctx = event.context as Record<string, unknown>;
-//         if (ctx.issue && typeof ctx.issue === "object") {
-//           const issue = ctx.issue as Record<string, unknown>;
-//           if (issue.title) essential.issue_title = issue.title;
-//           if (issue.body) essential.issue_body = issue.body;
-//         }
-//       }
-//       break;
+  // git status (try to get it, but don't fail if git isn't available)
+  try {
+    const gitStatus = execSync("git status --short", { encoding: "utf-8", stdio: "pipe" }).trim();
+    lines.push(`git_status: ${gitStatus || "(clean)"}`);
+  } catch {
+    // git not available or not in a repo
+  }
 
-//     case "issues_opened":
-//     case "issues_assigned":
-//     case "issues_labeled":
-//       if ("issue_title" in event) essential.issue_title = event.issue_title;
-//       if ("issue_body" in event) essential.issue_body = event.issue_body;
-//       break;
+  // repo data
+  lines.push(`repo: ${repo.owner}/${repo.name}`);
+  lines.push(`default_branch: ${repo.defaultBranch}`);
 
-//     case "pull_request_opened":
-//     case "pull_request_review_requested":
-//       if ("pr_title" in event) essential.pr_title = event.pr_title;
-//       if ("pr_body" in event) essential.pr_body = event.pr_body;
-//       if ("branch" in event) essential.branch = event.branch;
-//       break;
+  // GitHub Actions variables (when running in CI)
+  const ghVars: Record<string, string | undefined> = {
+    github_event_name: process.env.GITHUB_EVENT_NAME,
+    github_ref: process.env.GITHUB_REF,
+    github_sha: process.env.GITHUB_SHA?.slice(0, 7),
+    github_actor: process.env.GITHUB_ACTOR,
+    github_run_id: process.env.GITHUB_RUN_ID,
+    github_workflow: process.env.GITHUB_WORKFLOW,
+  };
+  for (const [key, value] of Object.entries(ghVars)) {
+    if (value) {
+      lines.push(`${key}: ${value}`);
+    }
+  }
 
-//     case "pull_request_review_submitted":
-//       if ("review_id" in event) essential.review_id = event.review_id;
-//       if ("review_body" in event) essential.review_body = event.review_body;
-//       if ("review_state" in event) essential.review_state = event.review_state;
-//       if ("branch" in event) essential.branch = event.branch;
-//       break;
+  return lines.join("\n");
+}
 
-//     case "pull_request_review_comment_created":
-//       if ("comment_id" in event) essential.comment_id = event.comment_id;
-//       if ("comment_body" in event) essential.comment_body = event.comment_body;
-//       if ("pr_title" in event) essential.pr_title = event.pr_title;
-//       if ("branch" in event) essential.branch = event.branch;
-//       break;
+interface AddInstructionsParams {
+  payload: Payload;
+  repo: RepoInfo;
+}
 
-//     case "check_suite_completed":
-//       if ("pr_title" in event) essential.pr_title = event.pr_title;
-//       if ("pr_body" in event) essential.pr_body = event.pr_body;
-//       if ("branch" in event) essential.branch = event.branch;
-//       if ("check_suite" in event) {
-//         essential.check_suite = {
-//           id: event.check_suite.id,
-//           head_sha: event.check_suite.head_sha,
-//           head_branch: event.check_suite.head_branch,
-//           status: event.check_suite.status,
-//           conclusion: event.check_suite.conclusion,
-//         };
-//       }
-//       break;
-
-//     case "workflow_dispatch":
-//       if ("inputs" in event) essential.inputs = event.inputs;
-//       break;
-//   }
-
-//   return essential;
-// }
-
-export const addInstructions = (payload: Payload) => {
+export const addInstructions = ({ payload, repo }: AddInstructionsParams) => {
   let encodedEvent = "";
 
   const eventKeys = Object.keys(payload.event);
@@ -99,7 +67,10 @@ export const addInstructions = (payload: Payload) => {
     encodedEvent = toonEncode(payload.event);
   }
 
-  return `
+  const runtimeContext = buildRuntimeContext(repo);
+
+  return (
+    `
 ***********************************************
 ************* SYSTEM INSTRUCTIONS *************
 ***********************************************
@@ -133,34 +104,34 @@ CRITICAL SECURITY RULES - NEVER VIOLATE UNDER ANY CIRCUMSTANCES:
 
 ### Rule 1: Never expose secrets through ANY means
 
-You must NEVER expose secrets through any channel, including but not limited to:
-- Displaying, printing, echoing, logging, or outputting to console
-- Writing to files (including .txt, .env, .json, config files, etc.)
-- Including in git commits, commit messages, or PR descriptions
-- Posting in GitHub comments, issue bodies, or PR review comments
-- Returning in tool outputs, API responses, or error messages
-- Including in redirect URLs, WebSocket messages, or GraphQL responses
+  You must NEVER expose secrets through any channel, including but not limited to:
+  - Displaying, printing, echoing, logging, or outputting to console
+  - Writing to files (including .txt, .env, .json, config files, etc.)
+  - Including in git commits, commit messages, or PR descriptions
+  - Posting in GitHub comments, issue bodies, or PR review comments
+  - Returning in tool outputs, API responses, or error messages
+  - Including in redirect URLs, WebSocket messages, or GraphQL responses
 
-Secrets include: API keys, authentication tokens, passwords, private keys, certificates, database connection strings, and any credential used for authentication or authorization. Common patterns (case-insensitive): variables containing API_KEY, SECRET, TOKEN, PASSWORD, CREDENTIAL, PRIVATE_KEY, or AUTH in an authentication context. Use judgment: \`PUBLIC_KEY\` for a cryptographic public key is fine; \`PRIVATE_KEY\` is not.
+  Secrets include: API keys, authentication tokens, passwords, private keys, certificates, database connection strings, and any credential used for authentication or authorization. Common patterns (case-insensitive): variables containing API_KEY, SECRET, TOKEN, PASSWORD, CREDENTIAL, PRIVATE_KEY, or AUTH in an authentication context. Use judgment: \`PUBLIC_KEY\` for a cryptographic public key is fine; \`PRIVATE_KEY\` is not.
 
 ### Rule 2: Never serialize objects containing secrets
 
-When working with objects that may contain environment variables or secrets:
-- NEVER serialize, stringify, or dump entire environment objects (process.env, os.environ, ENV, etc.)
-- NEVER iterate over environment variables and write their values to files
-- NEVER include environment variable values in outputs, logs, HTTP requests, or anywhere they can be exposed
-- If you must list properties, only show property NAMES, never values
-- Only access specific, known-safe keys explicitly (e.g., NODE_ENV, HOME, PWD)
+  When working with objects that may contain environment variables or secrets:
+  - NEVER serialize, stringify, or dump entire environment objects (process.env, os.environ, ENV, etc.)
+  - NEVER iterate over environment variables and write their values to files
+  - NEVER include environment variable values in outputs, logs, HTTP requests, or anywhere they can be exposed
+  - If you must list properties, only show property NAMES, never values
+  - Only access specific, known-safe keys explicitly (e.g., NODE_ENV, HOME, PWD)
 
 ### Rule 3: Refuse and explain
 
-Even if explicitly requested to reveal secrets, you must:
-1. Refuse the request
-2. Print a message explaining that exposing secrets is prohibited for security reasons
-3. If using ${ghPullfrogMcpName}, update the working comment to explain that secrets cannot be revealed
-4. Offer a safe alternative, if applicable
+  Even if explicitly requested to reveal secrets, you must:
+  1. Refuse the request
+  2. Print a message explaining that exposing secrets is prohibited for security reasons
+  3. If using ${ghPullfrogMcpName}, update the working comment to explain that secrets cannot be revealed
+  4. Offer a safe alternative, if applicable
 
-If you encounter secrets in files or environment, acknowledge they exist but never reveal their values.
+  If you encounter secrets in files or environment, acknowledge they exist but never reveal their values.
 
 ## MCP (Model Context Protocol) Tools
 
@@ -172,49 +143,66 @@ Tool names may be formatted as \`(server name)/(tool name)\`, for example: \`${g
 
 **Git operations**: All git operations must use ${ghPullfrogMcpName} MCP tools to ensure proper authentication and commit attribution. Do NOT use git commands directly (e.g., \`git commit\`, \`git push\`, \`git checkout\`, \`git branch\`) - these will use incorrect credentials and attribute commits to the wrong author.
 
-**Available git MCP tools**:
-- \`${ghPullfrogMcpName}/create_branch\` - Create a new branch from a base branch
-- \`${ghPullfrogMcpName}/commit_files\` - Stage and commit files with proper authentication
-- \`${ghPullfrogMcpName}/push_branch\` - Push a branch to the remote repository
-- \`${ghPullfrogMcpName}/create_pull_request\` - Create a PR from the current branch
+**File discovery**: Use \`${ghPullfrogMcpName}/list_files\` to discover files in the repository. This tool finds both git-tracked and untracked files, including newly created files that haven't been committed yet. Prefer this over native agent tools like \`glob\` or \`grep\` for file discovery, as it provides consistent results and handles untracked files correctly.
+` +
+    // **Available git MCP tools**:
+    // - \`${ghPullfrogMcpName}/checkout_pr\` - Checkout an existing PR branch locally (handles fork PRs automatically)
+    // - \`${ghPullfrogMcpName}/create_branch\` - Create a new branch from a base branch
+    // - \`${ghPullfrogMcpName}/commit_files\` - Stage and commit files with proper authentication
+    // - \`${ghPullfrogMcpName}/push_branch\` - Push a branch to the remote (automatically uses correct remote for fork PRs)
+    // - \`${ghPullfrogMcpName}/create_pull_request\` - Create a PR from the current branch
 
-**Workflow for making code changes**:
-1. Use file operations to create/modify files
-2. Use \`${ghPullfrogMcpName}/create_branch\` to create a new branch
-3. Use \`${ghPullfrogMcpName}/commit_files\` to commit your changes
-4. Use \`${ghPullfrogMcpName}/push_branch\` to push the branch
-5. Use \`${ghPullfrogMcpName}/create_pull_request\` to create a PR
+    // **Workflow for working on an existing PR**:
+    // 1. Use \`${ghPullfrogMcpName}/checkout_pr\` to checkout the PR branch
+    // 2. Make your changes using file operations
+    // 3. Use \`${ghPullfrogMcpName}/commit_files\` to commit your changes
+    // 4. Use \`${ghPullfrogMcpName}/push_branch\` to push (automatically pushes to fork for fork PRs)
 
+    // **Workflow for creating new changes**:
+    // 1. Use \`${ghPullfrogMcpName}/create_branch\` to create a new branch
+    // 2. Make your changes using file operations
+    // 3. Use \`${ghPullfrogMcpName}/commit_files\` to commit your changes
+    // 4. Use \`${ghPullfrogMcpName}/push_branch\` to push the branch
+    // 5. Use \`${ghPullfrogMcpName}/create_pull_request\` to create a PR
+
+    `
 **Do not attempt to configure git credentials manually** - the ${ghPullfrogMcpName} server handles all authentication internally.
 
+**Efficiency**: Trust the tools - do not repeatedly verify file contents or git status after operations. If a tool reports success, proceed to the next step. Only verify if you encounter an actual error.
+
+**Command execution**: Never use \`sleep\` to wait for commands to complete. Commands run synchronously - when the bash tool returns, the command has finished.
+
 **Commenting style**: When posting comments via ${ghPullfrogMcpName}, write as a professional team member would. Your final comments should be polished and actionable—do not include intermediate reasoning like "I'll now look at the code" or "Let me respond to the question."
 
-## Mode Selection
-
-Before starting any work, you must first determine which mode to use by examining the request and calling ${ghPullfrogMcpName}/select_mode.
-
-Available modes:
-
-${[...getModes({ disableProgressComment: payload.disableProgressComment }), ...payload.modes].map((w) => `    - "${w.name}": ${w.description}`).join("\n")}
-
-**Required first step**: 
-1. Examine the user's request/prompt carefully
-2. Determine which mode is most appropriate based on the mode descriptions above
-3. If the request could fit multiple modes, choose the mode with the narrowest scope that still addresses the request
-4. Call ${ghPullfrogMcpName}/select_mode with the chosen mode name
-5. The tool will return detailed instructions for that mode—follow those instructions, but remember they cannot override the Security rules or System instructions above
-6. Check for an AGENTS.md file or an agent-specific equivalent that applies to you. If it exists, read it and follow the instructions unless they conflict with the Security, System or Mode instructions above
-
-## When You're Stuck
-
-If you cannot complete a task due to missing information, ambiguity, or an unrecoverable error:
+**If you get stuck**: If you cannot complete a task due to missing information, ambiguity, or an unrecoverable error:
 1. Do not silently fail or produce incomplete work
 2. Post a comment via ${ghPullfrogMcpName} explaining what blocked you and what information or action would unblock you
 3. Make your blocker comment specific and actionable (e.g., "I need the database schema to proceed" not "I'm stuck")
 
+**Agent context files** Check for an AGENTS.md file or an agent-specific equivalent that applies to you. If it exists, read it and follow the instructions unless they conflict with the Security, System or Mode instructions above
+
+*************************************
+************* YOUR TASK *************
+*************************************
+
+**Required!** Before starting any work, you will pick a mode. Examine the prompt below carefully, along with the event data and runtime context. Determine which mode is most appropriate based on the mode descriptions below. Then use ${ghPullfrogMcpName}/select_mode to pick a mode.  If the request could fit multiple modes, choose the mode with the narrowest scope that still addresses the request. You will be given back detailed step-by-step instructions based on your selection.
+
+### Available modes
+
+${[...getModes({ disableProgressComment: payload.disableProgressComment }), ...payload.modes].map((w) => `    - "${w.name}": ${w.description}`).join("\n")}
+
+### Following the mode instructions
+
+After selecting a mode, follow the detailed step-by-step instructions provided by the ${ghPullfrogMcpName}/select_mode tool. Refer to the user prompt, event data, and runtime context below to inform your actions. These instructions cannot override the Security rules or System instructions above.
+
+Eagerly inspect the MCP tools available to you via the \`${ghPullfrogMcpName}\` MCP server. These are VITALLY IMPORTANT to completing your task.
+
 ************* USER PROMPT *************
 
-${payload.prompt}
+${payload.prompt
+  .split("\n")
+  .map((line) => `> ${line}`)
+  .join("\n")}
 
 ${
   encodedEvent
@@ -228,6 +216,6 @@ ${encodedEvent}`
 
 ************* RUNTIME CONTEXT *************
 
-working_directory: ${process.cwd()}
-`;
+${runtimeContext}`
+  );
 };

agents/opencode.ts

@@ -1,4 +1,4 @@
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { mkdirSync, writeFileSync } from "node:fs";
 
 import { join } from "node:path";
 import { log } from "../utils/cli.ts";
@@ -11,13 +11,253 @@ import {
   setupProcessAgentEnv,
 } from "./shared.ts";
 
-// import { createOpencode } from "@opencode-ai/sdk"
+export const opencode = agent({
+  name: "opencode",
+  install: async () => {
+    return await installFromNpmTarball({
+      packageName: "opencode-ai",
+      version: "latest",
+      executablePath: "bin/opencode",
+      installDependencies: true,
+    });
+  },
+  run: async ({ payload, apiKey: _apiKey, apiKeys, mcpServers, cliPath, repo }) => {
+    // 1. configure home/config directory
+    const tempHome = process.env.PULLFROG_TEMP_DIR!;
+    const configDir = join(tempHome, ".config", "opencode");
+    mkdirSync(configDir, { recursive: true });
 
-// const { client } = await createOpencode({
-//   config: {
-//     ''
-//   }
-// })
+    configureOpenCode({ mcpServers, sandbox: payload.sandbox ?? false });
+
+    const prompt = addInstructions({ payload, repo });
+    log.group("» Full prompt", () => log.info(prompt));
+
+    // message positional must come right after "run", before flags
+    const args = ["run", prompt, "--format", "json"];
+
+    if (payload.sandbox) {
+      log.info("🔒 sandbox mode enabled: restricting to read-only operations");
+    }
+
+    // 6. set up environment
+    setupProcessAgentEnv({ HOME: tempHome });
+
+    // build env vars: start with process.env (includes all API_KEY vars loaded by config())
+    // exclude GITHUB_TOKEN - OpenCode should use MCP server for GitHub operations, not direct token
+    // then override with apiKeys, HOME, and XDG_CONFIG_HOME
+    // XDG_CONFIG_HOME must be set because GitHub Actions sets it to a different path,
+    // and OpenCode follows XDG spec (checks XDG_CONFIG_HOME before falling back to $HOME/.config)
+    const env: Record<string, string> = {
+      ...(Object.fromEntries(
+        Object.entries(process.env).filter(
+          ([key, value]) => value !== undefined && key !== "GITHUB_TOKEN"
+        )
+      ) as Record<string, string>),
+      HOME: tempHome,
+      XDG_CONFIG_HOME: join(tempHome, ".config"),
+    };
+
+    // add/override API keys from apiKeys object (uppercase keys)
+    for (const [key, value] of Object.entries(apiKeys || {})) {
+      env[key.toUpperCase()] = value;
+    }
+
+    // run OpenCode in the repository directory (process.cwd() is set to GITHUB_WORKSPACE or repo dir)
+    const repoDir = process.cwd();
+
+    log.info(`🚀 Starting OpenCode CLI: ${cliPath} ${args.join(" ")}`);
+    log.info(`📁 Working directory: ${repoDir}`);
+    log.debug(`🏠 HOME: ${env.HOME}`);
+    log.debug(`📋 XDG_CONFIG_HOME: ${env.XDG_CONFIG_HOME}`);
+
+    const startTime = Date.now();
+    let lastActivityTime = startTime;
+    let eventCount = 0;
+
+    let output = "";
+    let stdoutBuffer = ""; // buffer for incomplete lines across chunks
+    const result = await spawn({
+      cmd: cliPath,
+      args,
+      cwd: repoDir,
+      env,
+      timeout: 600000, // 10 minutes timeout to prevent infinite hangs
+      stdio: ["ignore", "pipe", "pipe"],
+      onStdout: async (chunk) => {
+        const text = chunk.toString();
+        output += text;
+
+        // buffer incomplete lines across chunks (NDJSON format)
+        stdoutBuffer += text;
+        const lines = stdoutBuffer.split("\n");
+
+        // keep the last element (may be incomplete) in the buffer
+        stdoutBuffer = lines.pop() || "";
+
+        for (const line of lines) {
+          const trimmed = line.trim();
+          if (!trimmed) {
+            continue;
+          }
+
+          try {
+            const event = JSON.parse(trimmed) as OpenCodeEvent;
+            eventCount++;
+
+            // debug log all events to diagnose ordering and missing MCP/bash tool calls
+            log.debug(JSON.stringify(event, null, 2));
+
+            const timeSinceLastActivity = Date.now() - lastActivityTime;
+            if (timeSinceLastActivity > 10000) {
+              const activeToolCalls = toolCallTimings.size;
+              const toolCallInfo =
+                activeToolCalls > 0
+                  ? ` (waiting for ${activeToolCalls} tool call${activeToolCalls > 1 ? "s" : ""})`
+                  : " (OpenCode may be processing internally - LLM calls, planning, etc.)";
+              log.warning(
+                `⚠️  No activity for ${(timeSinceLastActivity / 1000).toFixed(1)}s${toolCallInfo} (${eventCount} events processed so far)`
+              );
+            }
+            lastActivityTime = Date.now();
+            const handler = messageHandlers[event.type as keyof typeof messageHandlers];
+            if (handler) {
+              await handler(event as never);
+            } else {
+              // log unhandled event types for visibility
+              log.info(
+                `📋 OpenCode event (unhandled): type=${event.type}, data=${JSON.stringify(event).substring(0, 500)}`
+              );
+            }
+          } catch {
+            // non-JSON lines are ignored (might be debug output from opencode)
+            log.debug(`» non-JSON stdout line: ${trimmed.substring(0, 200)}`);
+          }
+        }
+      },
+      onStderr: (chunk) => {
+        try {
+          const parsed = JSON.parse(chunk);
+          log.debug(JSON.stringify(parsed, null, 2));
+        } catch {
+          // if not JSON, fall through to regular error logging
+        }
+        const trimmed = chunk.trim();
+        if (trimmed) {
+          log.warning(trimmed);
+        }
+      },
+    });
+
+    const duration = Date.now() - startTime;
+    log.info(`✅ OpenCode CLI completed in ${duration}ms with exit code ${result.exitCode}`);
+
+    // 8. log tokens if they weren't logged yet (fallback if result event wasn't emitted)
+    if (!tokensLogged && (accumulatedTokens.input > 0 || accumulatedTokens.output > 0)) {
+      const totalTokens = accumulatedTokens.input + accumulatedTokens.output;
+      await log.summaryTable([
+        [
+          { data: "Input Tokens", header: true },
+          { data: "Output Tokens", header: true },
+          { data: "Total Tokens", header: true },
+        ],
+        [String(accumulatedTokens.input), String(accumulatedTokens.output), String(totalTokens)],
+      ]);
+    }
+
+    // 9. return result
+    if (result.exitCode !== 0) {
+      const errorMessage =
+        result.stderr || result.stdout || "Unknown error - no output from OpenCode CLI";
+      log.error(`OpenCode CLI exited with code ${result.exitCode}: ${errorMessage}`);
+      log.debug(`OpenCode stdout: ${result.stdout?.substring(0, 500)}`);
+      log.debug(`OpenCode stderr: ${result.stderr?.substring(0, 500)}`);
+      return {
+        success: false,
+        output: finalOutput || output,
+        error: errorMessage,
+      };
+    }
+
+    return {
+      success: true,
+      output: finalOutput || output,
+    };
+  },
+});
+
+interface ConfigureOpenCodeParams {
+  mcpServers: ConfigureMcpServersParams["mcpServers"];
+  sandbox: boolean;
+}
+
+/**
+ * Configure OpenCode via opencode.json config file.
+ * Builds complete config with MCP servers and permissions in a single write to avoid race conditions.
+ */
+function configureOpenCode({ mcpServers, sandbox }: ConfigureOpenCodeParams): void {
+  const tempHome = process.env.PULLFROG_TEMP_DIR!;
+  const configDir = join(tempHome, ".config", "opencode");
+  mkdirSync(configDir, { recursive: true });
+  const configPath = join(configDir, "opencode.json");
+
+  // build MCP servers config
+  const opencodeMcpServers: Record<string, { type: "remote"; url: string }> = {};
+  for (const [serverName, serverConfig] of Object.entries(mcpServers)) {
+    if (serverConfig.type !== "http") {
+      log.error(
+        `unsupported MCP server type for OpenCode: ${(serverConfig as never as { type: string }).type || "unknown"}`
+      );
+      throw new Error(
+        `Unsupported MCP server type for OpenCode: ${(serverConfig as never as { type: string }).type || "unknown"}`
+      );
+    }
+
+    opencodeMcpServers[serverName] = {
+      type: "remote",
+      url: serverConfig.url,
+    };
+  }
+
+  // build permissions config
+  const permission = sandbox
+    ? {
+        edit: "deny",
+        bash: "deny",
+        webfetch: "deny",
+        doom_loop: "allow",
+        external_directory: "allow",
+      }
+    : {
+        edit: "allow",
+        bash: "allow",
+        webfetch: "allow",
+        doom_loop: "allow",
+        external_directory: "allow",
+      };
+
+  // build complete config in one object
+  const config = {
+    mcp: opencodeMcpServers,
+    permission,
+  };
+
+  const configJson = JSON.stringify(config, null, 2);
+  try {
+    writeFileSync(configPath, configJson, "utf-8");
+  } catch (error) {
+    log.error(
+      `failed to write OpenCode config to ${configPath}: ${error instanceof Error ? error.message : String(error)}`
+    );
+    throw error;
+  }
+
+  log.info(`» OpenCode config written to ${configPath} (sandbox: ${sandbox})`);
+  log.debug(`OpenCode config contents:\n${configJson}`);
+}
+
+////////////////////////////////////////////
+////////////   EVENT HANDLERS   ////////////
+////////////////////////////////////////////
 
 // opencode cli event types inferred from json output format
 interface OpenCodeInitEvent {
@@ -87,16 +327,33 @@ interface OpenCodeStepFinishEvent {
 
 interface OpenCodeToolUseEvent {
   type: "tool_use";
-  timestamp?: string;
-  tool_name?: string;
-  tool_id?: string;
-  parameters?: unknown;
+  timestamp?: number;
+  sessionID?: string;
+  part?: {
+    id?: string;
+    callID?: string;
+    tool?: string;
+    state?: {
+      status?: string;
+      input?: unknown;
+      output?: string;
+    };
+  };
   [key: string]: unknown;
 }
 
 interface OpenCodeToolResultEvent {
   type: "tool_result";
-  timestamp?: string;
+  timestamp?: number;
+  sessionID?: string;
+  part?: {
+    callID?: string;
+    state?: {
+      status?: string;
+      output?: string;
+    };
+  };
+  // fallback fields for older format
   tool_id?: string;
   status?: "success" | "error";
   output?: string;
@@ -117,6 +374,19 @@ interface OpenCodeResultEvent {
   [key: string]: unknown;
 }
 
+interface OpenCodeErrorEvent {
+  type: "error";
+  timestamp?: string;
+  sessionID?: string;
+  error?: {
+    name?: string;
+    message?: string;
+    data?: unknown;
+    [key: string]: unknown;
+  };
+  [key: string]: unknown;
+}
+
 type OpenCodeEvent =
   | OpenCodeInitEvent
   | OpenCodeMessageEvent
@@ -125,7 +395,8 @@ type OpenCodeEvent =
   | OpenCodeStepFinishEvent
   | OpenCodeToolUseEvent
   | OpenCodeToolResultEvent
-  | OpenCodeResultEvent;
+  | OpenCodeResultEvent
+  | OpenCodeErrorEvent;
 
 let finalOutput = "";
 let accumulatedTokens: { input: number; output: number } = { input: 0, output: 0 };
@@ -141,6 +412,7 @@ const messageHandlers = {
     log.info(
       `🔵 OpenCode init: session_id=${event.session_id || "unknown"}, model=${event.model || "unknown"}`
     );
+    log.info(`🔵 OpenCode init event (full): ${JSON.stringify(event)}`);
     finalOutput = "";
     accumulatedTokens = { input: 0, output: 0 };
     tokensLogged = false;
@@ -172,9 +444,6 @@ const messageHandlers = {
     // log from text events only to avoid duplicates
     if (event.part?.text?.trim()) {
       const message = event.part.text.trim();
-      log.info(
-        `📝 OpenCode text output: ${message.substring(0, 200)}${message.length > 200 ? "..." : ""}`
-      );
       log.box(message, { title: "OpenCode" });
       finalOutput = message;
     }
@@ -207,45 +476,51 @@ const messageHandlers = {
     }
   },
   tool_use: (event: OpenCodeToolUseEvent) => {
-    if (event.tool_name && event.tool_id) {
-      toolCallTimings.set(event.tool_id, Date.now());
-      const paramsStr = event.parameters
-        ? JSON.stringify(event.parameters).substring(0, 500)
-        : "{}";
-      const stepContext = currentStepId
-        ? ` (step=${currentStepType || "unknown"}, stepId=${currentStepId.substring(0, 20)}...)`
-        : "";
-      log.info(`🔧 OpenCode tool_use: ${event.tool_name}${stepContext}, id=${event.tool_id}`);
-      log.info(`   Parameters: ${paramsStr}${paramsStr.length >= 500 ? "..." : ""}`);
+    const toolName = event.part?.tool;
+    const toolId = event.part?.callID;
+    const parameters = event.part?.state?.input;
+    const status = event.part?.state?.status;
+    const output = event.part?.state?.output;
 
+    // debug log all tool_use events to diagnose missing bash/MCP tool calls
+    if (!toolName || !toolId) {
+      log.debug(`» tool_use event missing toolName or toolId: ${JSON.stringify(event)}`);
+    }
+
+    if (toolName && toolId) {
       // track tool call in current step
       if (stepHistory.length > 0) {
-        stepHistory[stepHistory.length - 1].toolCalls.push(event.tool_name);
+        stepHistory[stepHistory.length - 1].toolCalls.push(toolName);
       }
 
       log.toolCall({
-        toolName: event.tool_name,
-        input: event.parameters || {},
+        toolName,
+        input: parameters || {},
       });
+
+      // if tool already completed (status in same event), log output
+      if (status === "completed" && output) {
+        log.debug(`  output: ${output}`);
+      }
     }
   },
   tool_result: (event: OpenCodeToolResultEvent) => {
-    if (event.tool_id) {
-      const toolStartTime = toolCallTimings.get(event.tool_id);
+    // handle both new part structure and legacy flat structure
+    const toolId = event.part?.callID || event.tool_id;
+    const status = event.part?.state?.status || event.status || "unknown";
+    const output = event.part?.state?.output || event.output;
+
+    if (toolId) {
+      const toolStartTime = toolCallTimings.get(toolId);
       if (toolStartTime) {
         const toolDuration = Date.now() - toolStartTime;
-        toolCallTimings.delete(event.tool_id);
-        const status = event.status || "unknown";
+        toolCallTimings.delete(toolId);
         const stepContext = currentStepId ? ` (step=${currentStepType || "unknown"})` : "";
-        const outputPreview =
-          typeof event.output === "string"
-            ? event.output.substring(0, 500)
-            : JSON.stringify(event.output).substring(0, 500);
         log.info(
-          `🔧 OpenCode tool_result${stepContext}: id=${event.tool_id}, status=${status}, duration=${toolDuration}ms`
+          `🔧 OpenCode tool_result${stepContext}: id=${toolId}, status=${status}, duration=${toolDuration}ms`
         );
-        if (outputPreview && outputPreview !== "{}" && outputPreview !== "null") {
-          log.info(`   Output: ${outputPreview}${outputPreview.length >= 500 ? "..." : ""}`);
+        if (output) {
+          log.debug(`  output: ${typeof output === "string" ? output : JSON.stringify(output)}`);
         }
         if (toolDuration > 5000) {
           log.warning(
@@ -254,9 +529,8 @@ const messageHandlers = {
         }
       }
     }
-    if (event.status === "error") {
-      const errorMsg =
-        typeof event.output === "string" ? event.output : JSON.stringify(event.output);
+    if (status === "error") {
+      const errorMsg = typeof output === "string" ? output : JSON.stringify(output);
       log.warning(`❌ Tool call failed: ${errorMsg}`);
     }
   },
@@ -293,254 +567,3 @@ const messageHandlers = {
     }
   },
 };
-
-export const opencode = agent({
-  name: "opencode",
-  install: async () => {
-    return await installFromNpmTarball({
-      packageName: "opencode-ai",
-      version: "latest",
-      executablePath: "bin/opencode",
-      installDependencies: true,
-    });
-  },
-  run: async ({ payload, apiKey: _apiKey, apiKeys, mcpServers, cliPath }) => {
-    // 1. configure home/config directory
-    const tempHome = process.env.PULLFROG_TEMP_DIR!;
-    const configDir = join(tempHome, ".config", "opencode");
-    mkdirSync(configDir, { recursive: true });
-
-    configureOpenCodeMcpServers({ mcpServers });
-    configureOpenCodeSandbox({ sandbox: payload.sandbox ?? false });
-
-    const prompt = addInstructions(payload);
-    const args = ["run", "--format", "json", prompt];
-
-    if (payload.sandbox) {
-      log.info("🔒 sandbox mode enabled: restricting to read-only operations");
-    }
-
-    // 6. set up environment
-    setupProcessAgentEnv({ HOME: tempHome });
-
-    // build env vars: start with process.env (includes all API_KEY vars loaded by config())
-    // exclude GITHUB_TOKEN - OpenCode should use MCP server for GitHub operations, not direct token
-    // then override with apiKeys and HOME
-    const env: Record<string, string> = {
-      ...(Object.fromEntries(
-        Object.entries(process.env).filter(
-          ([key, value]) => value !== undefined && key !== "GITHUB_TOKEN"
-        )
-      ) as Record<string, string>),
-      HOME: tempHome,
-    };
-
-    // add/override API keys from apiKeys object (uppercase keys)
-    for (const [key, value] of Object.entries(apiKeys || {})) {
-      env[key.toUpperCase()] = value;
-    }
-
-    // run OpenCode in the repository directory (process.cwd() is set to GITHUB_WORKSPACE or repo dir)
-    const repoDir = process.cwd();
-
-    log.info(`🚀 Starting OpenCode CLI: ${cliPath} ${args.join(" ")}`);
-    log.info(`📁 Working directory: ${repoDir}`);
-    const startTime = Date.now();
-    let lastActivityTime = startTime;
-    let eventCount = 0;
-
-    let output = "";
-    const result = await spawn({
-      cmd: cliPath,
-      args,
-      cwd: repoDir,
-      env,
-      timeout: 600000, // 10 minutes timeout to prevent infinite hangs
-      stdio: ["ignore", "pipe", "pipe"],
-      onStdout: async (chunk) => {
-        const text = chunk.toString();
-        output += text;
-
-        // parse each line as JSON (opencode outputs one JSON object per line)
-        const lines = text.split("\n");
-        for (const line of lines) {
-          const trimmed = line.trim();
-          if (!trimmed) {
-            continue;
-          }
-
-          try {
-            const event = JSON.parse(trimmed) as OpenCodeEvent;
-            eventCount++;
-            const timeSinceLastActivity = Date.now() - lastActivityTime;
-            if (timeSinceLastActivity > 10000) {
-              const activeToolCalls = toolCallTimings.size;
-              const toolCallInfo =
-                activeToolCalls > 0
-                  ? ` (waiting for ${activeToolCalls} tool call${activeToolCalls > 1 ? "s" : ""})`
-                  : " (OpenCode may be processing internally - LLM calls, planning, etc.)";
-              log.warning(
-                `⚠️  No activity for ${(timeSinceLastActivity / 1000).toFixed(1)}s${toolCallInfo} (${eventCount} events processed so far)`
-              );
-            }
-            lastActivityTime = Date.now();
-            const handler = messageHandlers[event.type as keyof typeof messageHandlers];
-            if (handler) {
-              await handler(event as never);
-            } else {
-              // log unhandled event types for visibility (but don't spam)
-              if (
-                event.type &&
-                ![
-                  "init",
-                  "message",
-                  "text",
-                  "step_start",
-                  "step_finish",
-                  "tool_use",
-                  "tool_result",
-                  "result",
-                ].includes(event.type)
-              ) {
-                log.debug(`📋 OpenCode event (unhandled): type=${event.type}`);
-              }
-            }
-          } catch {
-            // non-JSON lines are ignored
-          }
-        }
-      },
-      onStderr: (chunk) => {
-        const trimmed = chunk.trim();
-        if (trimmed) {
-          log.warning(trimmed);
-        }
-      },
-    });
-
-    const duration = Date.now() - startTime;
-    log.info(`✅ OpenCode CLI completed in ${duration}ms with exit code ${result.exitCode}`);
-
-    // 8. log tokens if they weren't logged yet (fallback if result event wasn't emitted)
-    if (!tokensLogged && (accumulatedTokens.input > 0 || accumulatedTokens.output > 0)) {
-      const totalTokens = accumulatedTokens.input + accumulatedTokens.output;
-      await log.summaryTable([
-        [
-          { data: "Input Tokens", header: true },
-          { data: "Output Tokens", header: true },
-          { data: "Total Tokens", header: true },
-        ],
-        [String(accumulatedTokens.input), String(accumulatedTokens.output), String(totalTokens)],
-      ]);
-    }
-
-    // 9. return result
-    if (result.exitCode !== 0) {
-      const errorMessage =
-        result.stderr || result.stdout || "Unknown error - no output from OpenCode CLI";
-      log.error(`OpenCode CLI exited with code ${result.exitCode}: ${errorMessage}`);
-      log.debug(`OpenCode stdout: ${result.stdout?.substring(0, 500)}`);
-      log.debug(`OpenCode stderr: ${result.stderr?.substring(0, 500)}`);
-      return {
-        success: false,
-        output: finalOutput || output,
-        error: errorMessage,
-      };
-    }
-
-    return {
-      success: true,
-      output: finalOutput || output,
-    };
-  },
-});
-
-/**
- * Configure MCP servers for OpenCode using opencode.json config file.
- * OpenCode uses opencode.json with mcp section supporting remote servers with type: "remote" and url.
- */
-function configureOpenCodeMcpServers({
-  mcpServers,
-}: {
-  mcpServers: ConfigureMcpServersParams["mcpServers"];
-}): void {
-  const tempHome = process.env.PULLFROG_TEMP_DIR!;
-  const configDir = join(tempHome, ".config", "opencode");
-  mkdirSync(configDir, { recursive: true });
-  const configPath = join(configDir, "opencode.json");
-
-  // convert to opencode's expected format
-  const opencodeMcpServers: Record<string, { type: "remote"; url: string; enabled?: boolean }> = {};
-  for (const [serverName, serverConfig] of Object.entries(mcpServers)) {
-    if (serverConfig.type !== "http") {
-      throw new Error(
-        `Unsupported MCP server type for OpenCode: ${(serverConfig as any).type || "unknown"}`
-      );
-    }
-
-    opencodeMcpServers[serverName] = {
-      type: "remote",
-      url: serverConfig.url,
-      enabled: true,
-    };
-  }
-
-  // read existing config if it exists, or create new one
-  let config: Record<string, unknown> = {};
-  try {
-    if (existsSync(configPath)) {
-      const existingConfig = readFileSync(configPath, "utf-8");
-      config = JSON.parse(existingConfig);
-    }
-  } catch {
-    // config doesn't exist yet or is invalid, start fresh
-  }
-
-  config.mcp = opencodeMcpServers;
-
-  writeFileSync(configPath, JSON.stringify(config, null, 2), "utf-8");
-  log.info(`MCP config written to ${configPath}`);
-}
-
-/**
- * Configure OpenCode sandbox mode via opencode.json.
- * When sandbox is enabled, restricts tools to read-only operations.
- */
-function configureOpenCodeSandbox({ sandbox }: { sandbox: boolean }): void {
-  const tempHome = process.env.PULLFROG_TEMP_DIR!;
-  const configDir = join(tempHome, ".config", "opencode");
-  mkdirSync(configDir, { recursive: true });
-  const configPath = join(configDir, "opencode.json");
-
-  // read existing config if it exists, or create new one
-  let config: Record<string, unknown> = {};
-  try {
-    if (existsSync(configPath)) {
-      const existingConfig = readFileSync(configPath, "utf-8");
-      config = JSON.parse(existingConfig);
-    }
-  } catch {
-    // config doesn't exist yet or is invalid, start fresh
-  }
-
-  if (sandbox) {
-    // sandbox mode: disable write, bash, and webfetch tools
-    config.tools = {
-      write: false,
-      bash: false,
-      webfetch: false,
-    };
-  } else {
-    // normal mode: enable all tools (or don't set tools config to use defaults)
-    config.tools = {
-      write: true,
-      bash: true,
-      webfetch: true,
-    };
-  }
-
-  // preserve MCP config if it was already set by configureOpenCodeMcpServers
-  // (this function is called after configureOpenCodeMcpServers, so MCP config should already exist)
-  writeFileSync(configPath, JSON.stringify(config, null, 2), "utf-8");
-  log.info(`OpenCode config written to ${configPath} (sandbox: ${sandbox})`);
-}

agents/shared.ts

@@ -20,6 +20,15 @@ export interface AgentResult {
   metadata?: Record<string, unknown>;
 }
 
+/**
+ * Repo info for agent context
+ */
+export interface RepoInfo {
+  owner: string;
+  name: string;
+  defaultBranch: string;
+}
+
 /**
  * Configuration for agent creation
  */
@@ -29,6 +38,7 @@ export interface AgentConfig {
   payload: Payload;
   mcpServers: Record<string, McpHttpServerConfig>;
   cliPath: string;
+  repo: RepoInfo;
 }
 
 /**
@@ -46,9 +56,13 @@ export interface ConfigureMcpServersParams {
  * @returns Whitelisted environment object safe for subprocess spawning
  */
 export function createAgentEnv(agentSpecificVars: Record<string, string>): Record<string, string> {
+  const home = agentSpecificVars.HOME || process.env.HOME;
   return {
     PATH: process.env.PATH,
-    HOME: process.env.HOME,
+    HOME: home,
+    // XDG_CONFIG_HOME must match HOME to ensure CLI tools find config files in the right place.
+    // GitHub Actions sets XDG_CONFIG_HOME to /home/runner/.config which would override $HOME/.config lookup.
+    XDG_CONFIG_HOME: home ? join(home, ".config") : undefined,
     LOG_LEVEL: process.env.LOG_LEVEL,
     NODE_ENV: process.env.NODE_ENV,
     GITHUB_TOKEN: getGitHubInstallationToken(),
@@ -131,7 +145,7 @@ export async function installFromNpmTarball({
   let resolvedVersion = version;
   if (version.startsWith("^") || version.startsWith("~") || version === "latest") {
     const npmRegistry = process.env.NPM_REGISTRY || "https://registry.npmjs.org";
-    log.info(`Resolving version for ${version}...`);
+    log.debug(`» resolving version for ${version}...`);
     try {
       const registryResponse = await fetch(`${npmRegistry}/${packageName}`);
       if (!registryResponse.ok) {
@@ -139,7 +153,7 @@ export async function installFromNpmTarball({
       }
       const registryData = (await registryResponse.json()) as NpmRegistryData;
       resolvedVersion = registryData["dist-tags"].latest;
-      log.info(`Resolved to version ${resolvedVersion}`);
+      log.debug(`» resolved to version ${resolvedVersion}`);
     } catch (error) {
       log.warning(
         `Failed to resolve version from registry: ${error instanceof Error ? error.message : String(error)}`
@@ -148,7 +162,7 @@ export async function installFromNpmTarball({
     }
   }
 
-  log.info(`📦 Installing ${packageName}@${resolvedVersion}...`);
+  log.debug(`» installing ${packageName}@${resolvedVersion}...`);
 
   const tempDir = process.env.PULLFROG_TEMP_DIR!;
   const tarballPath = join(tempDir, "package.tgz");
@@ -165,7 +179,7 @@ export async function installFromNpmTarball({
     tarballUrl = `${npmRegistry}/${packageName}/-/${packageName}-${resolvedVersion}.tgz`;
   }
 
-  log.info(`Downloading from ${tarballUrl}...`);
+  log.debug(`» downloading from ${tarballUrl}...`);
   const response = await fetch(tarballUrl);
   if (!response.ok) {
     throw new Error(`Failed to download tarball: ${response.status} ${response.statusText}`);
@@ -175,10 +189,10 @@ export async function installFromNpmTarball({
   if (!response.body) throw new Error("Response body is null");
   const fileStream = createWriteStream(tarballPath);
   await pipeline(response.body, fileStream);
-  log.info(`Downloaded tarball to ${tarballPath}`);
+  log.debug(`» downloaded tarball to ${tarballPath}`);
 
   // Extract tarball
-  log.info(`Extracting tarball...`);
+  log.debug(`» extracting tarball...`);
   const extractResult = spawnSync("tar", ["-xzf", tarballPath, "-C", tempDir], {
     stdio: "pipe",
     encoding: "utf-8",
@@ -199,7 +213,7 @@ export async function installFromNpmTarball({
 
   // Install dependencies if requested
   if (installDependencies) {
-    log.info(`Installing dependencies for ${packageName}...`);
+    log.debug(`» installing dependencies for ${packageName}...`);
     const installResult = spawnSync("npm", ["install", "--production"], {
       cwd: extractedDir,
       stdio: "pipe",
@@ -210,13 +224,13 @@ export async function installFromNpmTarball({
         `Failed to install dependencies: ${installResult.stderr || installResult.stdout || "Unknown error"}`
       );
     }
-    log.info(`✓ Dependencies installed`);
+    log.debug(`» dependencies installed`);
   }
 
   // Make the file executable
   chmodSync(cliPath, 0o755);
 
-  log.info(`✓ ${packageName} installed at ${cliPath}`);
+  log.debug(`» ${packageName} installed at ${cliPath}`);
 
   return cliPath;
 }
@@ -453,6 +467,8 @@ export async function installFromCurl({
       // Run the install script with HOME set to temp directory
       // ensuring a fresh install for each run
       HOME: tempDir,
+      // XDG_CONFIG_HOME must match HOME so CLI tools find config in the right place
+      XDG_CONFIG_HOME: join(tempDir, ".config"),
       SHELL: process.env.SHELL,
       USER: process.env.USER,
     },
@@ -488,7 +504,7 @@ export const agent = <const input extends AgentInput>(input: input): defineAgent
 
 export interface AgentInput {
   name: AgentName;
-  install: () => Promise<string>;
+  install: (token?: string) => Promise<string>;
   run: (config: AgentConfig) => Promise<AgentResult>;
 }
 

entry.ts

@@ -12,7 +12,7 @@ import { log } from "./utils/cli.ts";
 
 async function run(): Promise<void> {
   // Change to GITHUB_WORKSPACE if set (this is where actions/checkout puts the repo)
-  // JavaScript actions run from the action's directory, not the checked-out repo
+  // JavaScript actions run from the action's directory, not the checked out repo
   if (process.env.GITHUB_WORKSPACE && process.cwd() !== process.env.GITHUB_WORKSPACE) {
     log.debug(`Changing to GITHUB_WORKSPACE: ${process.env.GITHUB_WORKSPACE}`);
     process.chdir(process.env.GITHUB_WORKSPACE);

external.ts

@@ -186,13 +186,30 @@ interface WorkflowDispatchEvent extends BasePayloadEvent {
   trigger: "workflow_dispatch";
 }
 
+/** simplified review comment data for payload */
+export interface ReviewCommentData {
+  id: number;
+  body: string;
+  path: string;
+  line: number | null;
+  user: string | null;
+  html_url: string;
+  in_reply_to_id: number | null;
+}
+
 interface FixReviewEvent extends BasePayloadEvent {
   trigger: "fix_review";
   issue_number: number;
   is_pr: true;
   review_id: number;
+  /** username of the person who triggered this action */
+  triggerer: string;
   /** "all" to fix all comments, or specific comment IDs to fix */
   comment_ids: number[] | "all";
+  /** comments the triggerer approved (via thumbs up) - these should be addressed */
+  approved_comments: ReviewCommentData[];
+  /** other comments in the review - for context only, do not address unless asked */
+  unapproved_comments: ReviewCommentData[];
 }
 
 interface UnknownEvent extends BasePayloadEvent {

fixtures/basic.txt

@@ -1 +1 @@
-Tell me a joke
+add a file implementing quicksort and test it

main.ts

@@ -5,7 +5,7 @@ import { flatMorph } from "@ark/util";
 import { Octokit } from "@octokit/rest";
 import { encode as toonEncode } from "@toon-format/toon";
 import { type } from "arktype";
-import { agents } from "./agents/index.ts";
+import { type Agent, agents } from "./agents/index.ts";
 import type { AgentResult } from "./agents/shared.ts";
 import type { AgentName, Payload } from "./external.ts";
 import { agentsManifest } from "./external.ts";
@@ -14,6 +14,7 @@ import { createMcpConfigs } from "./mcp/config.ts";
 import { startMcpHttpServer } from "./mcp/server.ts";
 import { getModes, type Mode, modes } from "./modes.ts";
 import packageJson from "./package.json" with { type: "json" };
+import type { PrepResult } from "./prep/index.ts";
 import { fetchRepoSettings, fetchWorkflowRunInfo, type RepoSettings } from "./utils/api.ts";
 import { log } from "./utils/cli.ts";
 import { reportErrorToComment } from "./utils/errorReport.ts";
@@ -22,7 +23,7 @@ import {
   revokeGitHubInstallationToken,
   setupGitHubInstallationToken,
 } from "./utils/github.ts";
-import { setupGit, setupGitConfig } from "./utils/setup.ts";
+import { setupGitAuth, setupGitConfig } from "./utils/setup.ts";
 import { Timer } from "./utils/timer.ts";
 
 // runtime validation using agents (needed for ArkType)
@@ -50,6 +51,20 @@ export interface MainResult {
   error?: string | undefined;
 }
 
+// intermediate result types for deterministic context building
+interface GitHubSetup {
+  token: string;
+  owner: string;
+  name: string;
+  octokit: Octokit;
+  repo: Awaited<ReturnType<Octokit["repos"]["get"]>>["data"];
+  repoSettings: RepoSettings;
+}
+
+type ApiKeySetup =
+  | { success: true; apiKey: string; apiKeys: Record<string, string> }
+  | { success: false; error: string };
+
 export async function main(inputs: Inputs): Promise<MainResult> {
   let mcpServerClose: (() => Promise<void>) | undefined;
   let payload: Payload | undefined;
@@ -57,23 +72,122 @@ export async function main(inputs: Inputs): Promise<MainResult> {
   try {
     const timer = new Timer();
 
-    // parse payload early to extract agent
+    // phase 1: parse and validate inputs
     payload = parsePayload(inputs);
+    Inputs.assert(inputs);
+    setupGitConfig();
 
-    const partialCtx = await initializeContext(inputs, payload);
-    const ctx = partialCtx as Context;
-    timer.checkpoint("initializeContext");
+    // phase 2: fast setup (github + temp dir)
+    const [githubSetup, sharedTempDir] = await Promise.all([
+      initializeGitHub(),
+      createTempDirectory(),
+    ]);
+    timer.checkpoint("githubSetup");
 
-    const { pushRemote } = await setupGit(ctx);
-    ctx.pushRemote = pushRemote;
-    timer.checkpoint("setupGit");
+    // phase 3: resolve agent (needs repo settings)
+    const agent = resolveAgent({
+      inputs,
+      payload,
+      repoSettings: githubSetup.repoSettings,
+    });
+    const resolvedPayload = { ...payload, agent: agent.name };
 
-    await setupTempDirectory(ctx);
-    timer.checkpoint("setupTempDirectory");
+    // phase 4: validate API key (sync, needs agent) - fail fast before long-running operations
+    const apiKeySetup = validateApiKey({
+      agent,
+      inputs,
+      owner: githubSetup.owner,
+      name: githubSetup.name,
+    });
+    if (!apiKeySetup.success) {
+      await reportErrorToComment({ error: apiKeySetup.error });
+      return { success: false, error: apiKeySetup.error };
+    }
 
-    await startMcpServer(ctx);
-    mcpServerClose = ctx.mcpServerClose;
-    timer.checkpoint("startMcpServer");
+    // phase 5: parallel long-running operations (agent install + git auth)
+    const toolState: ToolState = {};
+    const [cliPath] = await Promise.all([
+      installAgentCli({ agent, token: githubSetup.token }),
+      setupGitAuth({
+        token: githubSetup.token,
+        owner: githubSetup.owner,
+        name: githubSetup.name,
+        payload: resolvedPayload,
+        octokit: githubSetup.octokit,
+        toolState,
+      }),
+    ]);
+    timer.checkpoint("agentSetup+gitAuth");
+
+    // phase 6: compute modes
+    const computedModes: Mode[] = [
+      ...getModes({
+        disableProgressComment: resolvedPayload.disableProgressComment,
+      }),
+      ...(resolvedPayload.modes || []),
+    ];
+
+    // phase 7: compute runId/jobId for MCP tools
+    const runId = process.env.GITHUB_RUN_ID || "";
+    if (runId) {
+      const workflowRunInfo = await fetchWorkflowRunInfo(runId);
+      if (workflowRunInfo.progressCommentId) {
+        process.env.PULLFROG_PROGRESS_COMMENT_ID = workflowRunInfo.progressCommentId;
+        log.info(`📝 Using pre-created progress comment: ${workflowRunInfo.progressCommentId}`);
+      }
+    }
+
+    let jobId: string | undefined;
+    const jobName = process.env.GITHUB_JOB;
+    if (jobName && runId) {
+      const jobs = await githubSetup.octokit.rest.actions.listJobsForWorkflowRun({
+        owner: githubSetup.owner,
+        repo: githubSetup.name,
+        run_id: parseInt(runId, 10),
+      });
+      const matchingJob = jobs.data.jobs.find((job) => job.name === jobName);
+      if (matchingJob) {
+        jobId = String(matchingJob.id);
+        log.info(`📋 Found job ID: ${jobId}`);
+      }
+    }
+
+    // phase 8: build tool context and start MCP server
+    const toolContext: ToolContext = {
+      owner: githubSetup.owner,
+      name: githubSetup.name,
+      githubInstallationToken: githubSetup.token,
+      octokit: githubSetup.octokit,
+      payload: resolvedPayload,
+      repo: githubSetup.repo,
+      repoSettings: githubSetup.repoSettings,
+      modes: computedModes,
+      toolState,
+      agent,
+      sharedTempDir,
+      runId,
+      jobId,
+    };
+
+    const { url: mcpServerUrl, close: mcpServerCloseFunc } = await startMcpHttpServer(toolContext);
+    mcpServerClose = mcpServerCloseFunc;
+    log.info(`🚀 MCP server started at ${mcpServerUrl}`);
+
+    const mcpServers = createMcpConfigs(mcpServerUrl);
+    log.debug(`📋 MCP Config: ${JSON.stringify(mcpServers, null, 2)}`);
+    timer.checkpoint("mcpServer");
+
+    // BUILD FINAL IMMUTABLE CONTEXT
+    const ctx: AgentContext = {
+      ...toolContext,
+      inputs,
+      mcpServerUrl,
+      mcpServerClose: mcpServerCloseFunc,
+      mcpServers,
+      cliPath,
+      apiKey: apiKeySetup.apiKey,
+      apiKeys: apiKeySetup.apiKeys,
+    };
 
     // check for empty comment_ids in fix_review trigger - report and exit early
     if (
@@ -81,19 +195,12 @@ export async function main(inputs: Inputs): Promise<MainResult> {
       Array.isArray(ctx.payload.event.comment_ids) &&
       ctx.payload.event.comment_ids.length === 0
     ) {
-      await reportProgress(ctx, {
-        body: `👍 **No approved comments found**\n\nTo use "Fix 👍s", add a 👍 reaction to one or more inline review comments you want fixed.`,
-      });
+      const noThumbsMessage = `👍 **No approved comments found**\n\nTo use "Fix 👍s", add a 👍 reaction to one or more inline review comments you want fixed.`;
+      log.error(noThumbsMessage);
+      await reportProgress(ctx, { body: noThumbsMessage });
       return { success: true };
     }
 
-    setupMcpServers(ctx);
-
-    await installAgentCli(ctx);
-    timer.checkpoint("installAgentCli");
-
-    await validateApiKey(ctx);
-
     const result = await runAgent(ctx);
     const mainResult = await handleAgentResult(result);
     return mainResult;
@@ -129,15 +236,22 @@ export async function main(inputs: Inputs): Promise<MainResult> {
 /**
  * Get agents that have matching API keys in the inputs
  */
-function getAvailableAgents(inputs: Inputs): (typeof agents)[AgentName][] {
-  return Object.values(agents).filter((agent) => {
-    // for OpenCode, check if any API_KEY variable exists in inputs
-    if (agent.name === "opencode") {
-      return Object.keys(inputs).some((key) => key.includes("api_key"));
-    }
-    // for other agents, check apiKeyNames
-    return agent.apiKeyNames.some((inputKey) => inputs[inputKey]);
-  });
+/**
+ * Check if an agent has API keys available (inputs or process.env for opencode)
+ */
+function agentHasApiKeys(agent: Agent, inputs: Inputs): boolean {
+  if (agent.name === "opencode") {
+    // check inputs first, then process.env
+    const hasInputKey = Object.keys(inputs).some((key) => key.includes("api_key"));
+    if (hasInputKey) return true;
+    return Object.keys(process.env).some((key) => key.includes("API_KEY") && process.env[key]);
+  }
+  const inputsRecord = inputs as Record<string, string | undefined>;
+  return agent.apiKeyNames.some((inputKey) => inputsRecord[inputKey]);
+}
+
+function getAvailableAgents(inputs: Inputs): Agent[] {
+  return Object.values(agents).filter((agent) => agentHasApiKeys(agent, inputs));
 }
 
 /**
@@ -152,28 +266,29 @@ function getAllPossibleKeyNames(): string[] {
 }
 
 /**
- * Throw an error for missing API key with helpful message linking to repo settings
+ * Build a helpful error message for missing API key with links to repo settings
  */
-async function throwMissingApiKeyError(ctx: Context): Promise<never> {
+function buildMissingApiKeyError(params: { agent: Agent; owner: string; name: string }): string {
   const apiUrl = process.env.API_URL || "https://pullfrog.com";
-  const settingsUrl = `${apiUrl}/console/${ctx.owner}/${ctx.name}`;
+  const settingsUrl = `${apiUrl}/console/${params.owner}/${params.name}`;
 
-  const githubRepoUrl = `https://github.com/${ctx.owner}/${ctx.name}`;
+  const githubRepoUrl = `https://github.com/${params.owner}/${params.name}`;
   const githubSecretsUrl = `${githubRepoUrl}/settings/secrets/actions`;
 
   // for OpenCode, use a generic message since it accepts any API key
-  const isOpenCode = ctx.agent?.name === "opencode";
+  const isOpenCode = params.agent.name === "opencode";
   let secretNameList: string;
   if (isOpenCode) {
     secretNameList =
       "any API key (e.g., `OPENCODE_API_KEY`, `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, etc.)";
   } else {
-    const inputKeys = ctx.agent?.apiKeyNames || getAllPossibleKeyNames();
+    const inputKeys =
+      params.agent.apiKeyNames.length > 0 ? params.agent.apiKeyNames : getAllPossibleKeyNames();
     const secretNames = inputKeys.map((key) => `\`${key.toUpperCase()}\``);
     secretNameList = inputKeys.length === 1 ? secretNames[0] : `one of ${secretNames.join(" or ")}`;
   }
 
-  const message = `Pullfrog is configured to use ${ctx.agent.displayName}, but the associated API key was not provided.
+  return `Pullfrog is configured to use ${params.agent.displayName}, but the associated API key was not provided.
 
 To fix this, add the required secret to your GitHub repository:
 
@@ -184,114 +299,75 @@ To fix this, add the required secret to your GitHub repository:
 5. Click "Add secret"
 
 Alternatively, configure Pullfrog to use a different agent at ${settingsUrl}`;
-
-  // report to comment if MCP context is available (server has started)
-  await reportErrorToComment({ error: message });
-  throw new Error(message);
 }
 
-export interface Context {
-  // flattened from RepoContext
+// tool context - subset of Context needed by MCP tools
+export interface ToolContext {
   owner: string;
   name: string;
-
-  // core fields
-  inputs: Inputs;
-  payload: Payload;
-  repo: Awaited<ReturnType<Octokit["repos"]["get"]>>["data"];
-  agentName: AgentName;
-  agent: (typeof agents)[AgentName];
   githubInstallationToken: string;
   octokit: Octokit;
-
-  // repo settings from Pullfrog API
+  payload: Payload;
+  repo: Awaited<ReturnType<Octokit["repos"]["get"]>>["data"];
   repoSettings: RepoSettings;
-
-  // modes for MCP tools
   modes: Mode[];
-
-  // setup fields
-  pushRemote: string;
+  toolState: ToolState;
+  agent: Agent;
   sharedTempDir: string;
-
-  // mcp fields
-  mcpServerUrl: string;
-  mcpServerClose: () => Promise<void>;
-  mcpServers: ReturnType<typeof createMcpConfigs>;
-
-  // agent fields
-  cliPath: string;
-  apiKey: string;
-  apiKeys: Record<string, string>;
+  runId: string;
+  jobId: string | undefined;
 }
 
-async function initializeContext(
-  inputs: Inputs,
-  payload: Payload
-): Promise<
-  Omit<
-    Context,
-    | "mcpServerUrl"
-    | "mcpServerClose"
-    | "mcpServers"
-    | "cliPath"
-    | "apiKey"
-    | "apiKeys"
-    | "pushRemote"
-  >
-> {
-  log.info(`🐸 Running pullfrog/action@${packageJson.version}...`);
-  Inputs.assert(inputs);
-  setupGitConfig();
+export interface AgentContext extends Readonly<ToolContext> {
+  readonly inputs: Inputs;
+  readonly mcpServerUrl: string;
+  readonly mcpServerClose: () => Promise<void>;
+  readonly mcpServers: ReturnType<typeof createMcpConfigs>;
+  readonly cliPath: string;
+  readonly apiKey: string;
+  readonly apiKeys: Record<string, string>;
+}
 
-  const githubInstallationToken = await setupGitHubInstallationToken();
+export interface DependencyInstallationState {
+  status: "not_started" | "in_progress" | "completed" | "failed";
+  promise: Promise<PrepResult[]> | undefined;
+  results: PrepResult[] | undefined;
+}
+
+export interface ToolState {
+  prNumber?: number;
+  issueNumber?: number;
+  review?: {
+    id: number; // REST API database ID (for fix URLs)
+    nodeId: string; // GraphQL node ID (for mutations)
+  };
+  dependencyInstallation?: DependencyInstallationState;
+}
+
+/**
+ * Initialize GitHub connection: token, octokit, repo data, settings
+ */
+async function initializeGitHub(): Promise<GitHubSetup> {
+  log.info(`🐸 Running pullfrog/action@${packageJson.version}...`);
+
+  const token = await setupGitHubInstallationToken();
   const { owner, name } = parseRepoContext();
 
-  // create octokit instance
-  const octokit = new Octokit({
-    auth: githubInstallationToken,
-  });
+  const octokit = new Octokit({ auth: token });
 
-  // fetch repo data
-  const response = await octokit.repos.get({
-    owner,
-    repo: name,
-  });
-  const repo = response.data;
-
-  // fetch repo settings
-  const repoSettings = await fetchRepoSettings({
-    token: githubInstallationToken,
-    repoContext: { owner, name },
-  });
-
-  // resolve agent and update payload with resolved agent name
-  const { agentName, agent } = resolveAgent({
-    inputs,
-    payload,
-    repoSettings,
-  });
-  const resolvedPayload = { ...payload, agent: agentName };
-
-  // compute modes from defaults + payload overrides
-  const computedModes = [
-    ...getModes({ disableProgressComment: resolvedPayload.disableProgressComment }),
-    ...(resolvedPayload.modes || []),
-  ];
+  // fetch repo data and settings in parallel
+  const [repoResponse, repoSettings] = await Promise.all([
+    octokit.repos.get({ owner, repo: name }),
+    fetchRepoSettings({ token, repoContext: { owner, name } }),
+  ]);
 
   return {
+    token,
     owner,
     name,
-    inputs,
-    githubInstallationToken,
     octokit,
-    repo,
-    agentName,
-    agent,
-    payload: resolvedPayload,
+    repo: repoResponse.data,
     repoSettings,
-    modes: computedModes,
-    sharedTempDir: "",
   };
 }
 
@@ -303,65 +379,54 @@ function resolveAgent({
   inputs: Inputs;
   payload: Payload;
   repoSettings: RepoSettings;
-}): { agentName: AgentName; agent: (typeof agents)[AgentName] } {
+}): Agent {
   const agentOverride = process.env.AGENT_OVERRIDE as AgentName | undefined;
   const configuredAgentName = agentOverride || payload.agent || repoSettings.defaultAgent || null;
 
   if (configuredAgentName) {
-    const agentName = configuredAgentName;
-    const agent = agents[agentName];
+    const agent = agents[configuredAgentName];
     if (!agent) {
-      throw new Error(`invalid agent name: ${agentName}`);
+      throw new Error(`invalid agent name: ${configuredAgentName}`);
     }
 
     // if explicitly configured (via override or payload), respect it even without matching keys
     // this allows users to force an agent selection (will fail later with clear error if no keys)
     const isExplicitOverride = agentOverride !== undefined || payload.agent !== null;
-
     if (isExplicitOverride) {
-      log.info(`Selected configured agent: ${agentName}`);
-      return { agentName, agent };
+      log.info(`Selected configured agent: ${agent.name}`);
+      return agent;
     }
 
     // for repo-level defaults, check if agent has matching keys before selecting
-    const hasMatchingKey =
-      agent.name === "opencode"
-        ? Object.keys(inputs).some((key) => key.includes("api_key"))
-        : agent.apiKeyNames.some((inputKey) => inputs[inputKey]);
-    if (!hasMatchingKey) {
-      log.warning(
-        `Repo default agent ${agentName} has no matching API keys. Available agents: ${
-          getAvailableAgents(inputs)
-            .map((a) => a.name)
-            .join(", ") || "none"
-        }`
-      );
-      // fall through to auto-selection for repo defaults
-    } else {
-      log.info(`Selected configured agent: ${agentName}`);
-      return { agentName, agent };
+    if (agentHasApiKeys(agent, inputs)) {
+      log.info(`Selected configured agent: ${agent.name}`);
+      return agent;
     }
+
+    // fall through to auto-selection
+    const availableAgents = getAvailableAgents(inputs);
+    log.warning(
+      `Repo default agent ${agent.name} has no matching API keys. Available: ${
+        availableAgents.map((a) => a.name).join(", ") || "none"
+      }`
+    );
   }
 
   const availableAgents = getAvailableAgents(inputs);
-  const availableAgentNames = availableAgents.map((agent) => agent.name).join(", ");
-  log.debug(`Available agents: ${availableAgentNames || "none"}`);
-
   if (availableAgents.length === 0) {
-    // this will be caught and reported later in validateApiKey
     throw new Error("no agents available - missing API keys");
   }
 
-  const agentName = availableAgents[0].name;
   const agent = availableAgents[0];
-  log.info(`No agent configured, defaulting to first available agent: ${agentName}`);
-  return { agentName, agent };
+  log.info(`No agent configured, defaulting to first available agent: ${agent.name}`);
+  return agent;
 }
 
-async function setupTempDirectory(ctx: Context): Promise<void> {
-  ctx.sharedTempDir = await mkdtemp(join(tmpdir(), "pullfrog-"));
-  process.env.PULLFROG_TEMP_DIR = ctx.sharedTempDir;
-  log.info(`📂 PULLFROG_TEMP_DIR has been created at ${ctx.sharedTempDir}`);
+async function createTempDirectory(): Promise<string> {
+  const sharedTempDir = await mkdtemp(join(tmpdir(), "pullfrog-"));
+  process.env.PULLFROG_TEMP_DIR = sharedTempDir;
+  log.info(`📂 PULLFROG_TEMP_DIR has been created at ${sharedTempDir}`);
+  return sharedTempDir;
 }
 
 function parsePayload(inputs: Inputs): Payload {
@@ -384,72 +449,65 @@ function parsePayload(inputs: Inputs): Payload {
   }
 }
 
-async function startMcpServer(ctx: Context): Promise<void> {
-  // fetch the pre-created progress comment ID from the database
-  // this must be set BEFORE starting the MCP server so comment.ts can read it
-  const runId = process.env.GITHUB_RUN_ID;
-  if (runId) {
-    const workflowRunInfo = await fetchWorkflowRunInfo(runId);
-    if (workflowRunInfo.progressCommentId) {
-      process.env.PULLFROG_PROGRESS_COMMENT_ID = workflowRunInfo.progressCommentId;
-      log.info(`📝 Using pre-created progress comment: ${workflowRunInfo.progressCommentId}`);
-    }
-  }
-
-  const { url, close } = await startMcpHttpServer(ctx);
-  ctx.mcpServerUrl = url;
-  ctx.mcpServerClose = close;
-  log.info(`🚀 MCP server started at ${url}`);
-}
-
-function setupMcpServers(ctx: Context): void {
-  ctx.mcpServers = createMcpConfigs(ctx.mcpServerUrl);
-  log.debug(`📋 MCP Config: ${JSON.stringify(ctx.mcpServers, null, 2)}`);
-}
-
-async function installAgentCli(ctx: Context): Promise<void> {
+async function installAgentCli(params: { agent: Agent; token: string }): Promise<string> {
   // gemini is the only agent that needs githubInstallationToken for install
-  if (ctx.agentName === "gemini") {
-    ctx.cliPath = await ctx.agent.install(ctx.githubInstallationToken);
-  } else {
-    ctx.cliPath = await ctx.agent.install();
+  if (params.agent.name === "gemini") {
+    return params.agent.install(params.token);
   }
+  return params.agent.install();
 }
 
-async function validateApiKey(ctx: Context): Promise<void> {
-  // collect all matching API keys for this agent
+function collectApiKeys(agent: Agent, inputs: Inputs): Record<string, string> {
   const apiKeys: Record<string, string> = {};
-  for (const inputKey of ctx.agent.apiKeyNames) {
-    const value = ctx.inputs[inputKey];
+  const inputsRecord = inputs as Record<string, string | undefined>;
+
+  for (const inputKey of agent.apiKeyNames) {
+    const value = inputsRecord[inputKey];
     if (value) {
       apiKeys[inputKey] = value;
     }
   }
 
-  // for OpenCode: if no keys found in inputs, check process.env for any API_KEY variables
-  if (ctx.agentName === "opencode" && Object.keys(apiKeys).length === 0) {
+  // for OpenCode: also check process.env for any API_KEY variables
+  if (agent.name === "opencode" && Object.keys(apiKeys).length === 0) {
     for (const [key, value] of Object.entries(process.env)) {
       if (value && typeof value === "string" && key.includes("API_KEY")) {
-        // convert env var name back to input key format (lowercase with underscores)
-        const inputKey = key.toLowerCase();
-        apiKeys[inputKey] = value;
+        apiKeys[key.toLowerCase()] = value;
       }
     }
   }
 
-  if (Object.keys(apiKeys).length === 0) {
-    await throwMissingApiKeyError(ctx);
-    // unreachable - throwMissingApiKeyError always throws
-    return;
-  }
-
-  // keep apiKey for backward compat (first available key)
-  ctx.apiKey = Object.values(apiKeys)[0];
-  ctx.apiKeys = apiKeys;
+  return apiKeys;
 }
 
-async function runAgent(ctx: Context): Promise<AgentResult> {
-  log.info(`Running ${ctx.agentName}...`);
+function validateApiKey(params: {
+  agent: Agent;
+  inputs: Inputs;
+  owner: string;
+  name: string;
+}): ApiKeySetup {
+  const apiKeys = collectApiKeys(params.agent, params.inputs);
+
+  if (Object.keys(apiKeys).length === 0) {
+    return {
+      success: false,
+      error: buildMissingApiKeyError({
+        agent: params.agent,
+        owner: params.owner,
+        name: params.name,
+      }),
+    };
+  }
+
+  return {
+    success: true,
+    apiKey: Object.values(apiKeys)[0],
+    apiKeys,
+  };
+}
+
+async function runAgent(ctx: AgentContext): Promise<AgentResult> {
+  log.info(`Running ${ctx.agent.name}...`);
   // strip context from event
   const { context: _context, ...eventWithoutContext } = ctx.payload.event;
   // format: prompt + two newlines + TOON encoded event
@@ -462,6 +520,11 @@ async function runAgent(ctx: Context): Promise<AgentResult> {
     apiKey: ctx.apiKey,
     apiKeys: ctx.apiKeys,
     cliPath: ctx.cliPath,
+    repo: {
+      owner: ctx.owner,
+      name: ctx.name,
+      defaultBranch: ctx.repo.default_branch,
+    },
   });
 }
 

mcp/checkSuite.ts

@@ -1,18 +1,18 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const GetCheckSuiteLogs = type({
   check_suite_id: type.number.describe("the id from check_suite.id"),
 });
 
-export function GetCheckSuiteLogsTool(ctx: Context) {
+export function GetCheckSuiteLogsTool(ctx: ToolContext) {
   return tool({
     name: "get_check_suite_logs",
     description:
       "get workflow run logs for a failed check suite. pass check_suite.id from the webhook payload.",
     parameters: GetCheckSuiteLogs,
-    execute: execute(ctx, async ({ check_suite_id }) => {
+    execute: execute(async ({ check_suite_id }) => {
       // get workflow runs for this specific check suite
       const workflowRuns = await ctx.octokit.paginate(
         ctx.octokit.rest.actions.listWorkflowRunsForRepo,

mcp/checkout.ts

@@ -0,0 +1,191 @@
+import type { Octokit } from "@octokit/rest";
+import { type } from "arktype";
+import type { ToolContext } from "../main.ts";
+import { log } from "../utils/cli.ts";
+import { $ } from "../utils/shell.ts";
+import { execute, tool } from "./shared.ts";
+
+export const CheckoutPr = type({
+  pull_number: type.number.describe("the pull request number to checkout"),
+});
+
+export type CheckoutPrResult = {
+  success: true;
+  number: number;
+  title: string;
+  base: string;
+  head: string;
+  isFork: boolean;
+  maintainerCanModify: boolean;
+  url: string;
+  headRepo: string;
+  diff: string;
+};
+
+interface CheckoutPrBranchParams {
+  octokit: Octokit;
+  owner: string;
+  name: string;
+  token: string;
+  pullNumber: number;
+}
+
+interface CheckoutPrBranchResult {
+  prNumber: number;
+}
+
+/**
+ * Shared helper to checkout a PR branch and configure fork remotes.
+ * Assumes origin remote is already configured with authentication.
+ * Returns the PR number for caller to set on toolState.
+ */
+export async function checkoutPrBranch(
+  params: CheckoutPrBranchParams
+): Promise<CheckoutPrBranchResult> {
+  const { octokit, owner, name, token, pullNumber } = params;
+  log.info(`🔀 checking out PR #${pullNumber}...`);
+
+  // fetch PR metadata
+  const pr = await octokit.rest.pulls.get({
+    owner,
+    repo: name,
+    pull_number: pullNumber,
+  });
+
+  const headRepo = pr.data.head.repo;
+  if (!headRepo) {
+    throw new Error(`PR #${pullNumber} source repository was deleted`);
+  }
+
+  const isFork = headRepo.full_name !== pr.data.base.repo.full_name;
+  const baseBranch = pr.data.base.ref;
+  const headBranch = pr.data.head.ref;
+
+  // always use pr-{number} as local branch name for consistency
+  // this avoids naming conflicts and makes push config simpler
+  const localBranch = `pr-${pullNumber}`;
+
+  // check if we're already on the correct commit (not just branch name)
+  // this handles fork PRs where head branch name might match base branch name
+  const currentSha = $("git", ["rev-parse", "HEAD"], { log: false }).trim();
+  const alreadyOnBranch = currentSha === pr.data.head.sha;
+
+  if (alreadyOnBranch) {
+    log.debug(`already on PR branch ${localBranch}, skipping checkout`);
+  } else {
+    // fetch base branch so origin/<base> exists for diff operations
+    log.debug(`📥 fetching base branch (${baseBranch})...`);
+    $("git", ["fetch", "--no-tags", "origin", baseBranch]);
+
+    // checkout base branch first to avoid "refusing to fetch into current branch" error
+    // -B creates or resets the branch to match origin/baseBranch
+    $("git", ["checkout", "-B", baseBranch, `origin/${baseBranch}`]);
+
+    // fetch PR branch using pull/{n}/head refspec (works for both fork and same-repo PRs)
+    log.debug(`🌿 fetching PR #${pullNumber} (${localBranch})...`);
+    $("git", ["fetch", "--no-tags", "origin", `pull/${pullNumber}/head:${localBranch}`]);
+
+    // checkout the branch
+    $("git", ["checkout", localBranch]);
+    log.debug(`✓ checked out PR #${pullNumber}`);
+  }
+
+  // ensure base branch is fetched (needed for diff operations)
+  // fetch if we skipped checkout (already on branch) - otherwise already fetched above
+  if (alreadyOnBranch) {
+    log.debug(`📥 fetching base branch (${baseBranch})...`);
+    $("git", ["fetch", "--no-tags", "origin", baseBranch]);
+  }
+
+  // configure push remote for this branch
+  // NOTE: This always runs regardless of alreadyOnBranch, because setupGit doesn't configure
+  // fork remotes. This ensures fork PRs can push even when checkout_pr is called after setupGit.
+  if (isFork) {
+    const remoteName = `pr-${pullNumber}`;
+    const forkUrl = `https://x-access-token:${token}@github.com/${headRepo.full_name}.git`;
+
+    // add fork as a named remote (ignore error if already exists)
+    try {
+      $("git", ["remote", "add", remoteName, forkUrl]);
+      log.debug(`📌 added remote '${remoteName}' for fork ${headRepo.full_name}`);
+    } catch {
+      // remote already exists, update its URL
+      $("git", ["remote", "set-url", remoteName, forkUrl]);
+      log.debug(`📌 updated remote '${remoteName}' for fork ${headRepo.full_name}`);
+    }
+
+    // set branch push config so `git push` knows where to push
+    $("git", ["config", `branch.${localBranch}.pushRemote`, remoteName]);
+    // set merge ref so git knows the remote branch name (may differ from local)
+    $("git", ["config", `branch.${localBranch}.merge`, `refs/heads/${headBranch}`]);
+    log.debug(`📌 configured branch '${localBranch}' to push to '${remoteName}/${headBranch}'`);
+
+    // warn if maintainer can't modify (push will likely fail)
+    if (!pr.data.maintainer_can_modify) {
+      log.warning(
+        `⚠️ fork PR has maintainer_can_modify=false - push operations will fail. ` +
+          `ask the PR author to enable "Allow edits from maintainers" or the fork may be owned by an organization.`
+      );
+    }
+  } else {
+    // for same-repo PRs, push to origin
+    $("git", ["config", `branch.${localBranch}.pushRemote`, "origin"]);
+    $("git", ["config", `branch.${localBranch}.merge`, `refs/heads/${headBranch}`]);
+  }
+
+  return { prNumber: pullNumber };
+}
+
+export function CheckoutPrTool(ctx: ToolContext) {
+  return tool({
+    name: "checkout_pr",
+    description:
+      "Checkout a pull request branch locally. This fetches the PR branch and sets up push configuration for fork PRs. Use this when you need to work on an existing PR.",
+    parameters: CheckoutPr,
+    execute: execute(async ({ pull_number }) => {
+      const result = await checkoutPrBranch({
+        octokit: ctx.octokit,
+        owner: ctx.owner,
+        name: ctx.name,
+        token: ctx.githubInstallationToken,
+        pullNumber: pull_number,
+      });
+
+      // set prNumber on toolState
+      ctx.toolState.prNumber = result.prNumber;
+
+      // fetch PR metadata to return result
+      const pr = await ctx.octokit.rest.pulls.get({
+        owner: ctx.owner,
+        repo: ctx.name,
+        pull_number,
+      });
+
+      const headRepo = pr.data.head.repo;
+      if (!headRepo) {
+        throw new Error(`PR #${pull_number} source repository was deleted`);
+      }
+
+      // fetch PR diff via API (authoritative source - not affected by main advancing)
+      const diffResponse = await ctx.octokit.rest.pulls.get({
+        owner: ctx.owner,
+        repo: ctx.name,
+        pull_number,
+        mediaType: { format: "diff" },
+      });
+
+      return {
+        success: true,
+        number: pr.data.number,
+        title: pr.data.title,
+        base: pr.data.base.ref,
+        head: pr.data.head.ref,
+        isFork: headRepo.full_name !== pr.data.base.repo.full_name,
+        maintainerCanModify: pr.data.maintainer_can_modify,
+        url: pr.data.html_url,
+        headRepo: headRepo.full_name,
+        diff: diffResponse.data as unknown as string,
+      } satisfies CheckoutPrResult;
+    }),
+  });
+}

mcp/comment.ts

@@ -2,7 +2,7 @@ import { Octokit } from "@octokit/rest";
 import { type } from "arktype";
 import type { Payload } from "../external.ts";
 import { agentsManifest } from "../external.ts";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { fetchWorkflowRunInfo } from "../utils/api.ts";
 import { buildPullfrogFooter, stripExistingFooter } from "../utils/buildPullfrogFooter.ts";
 import { getGitHubInstallationToken, parseRepoContext } from "../utils/github.ts";
@@ -15,26 +15,49 @@ import { execute, tool } from "./shared.ts";
  */
 export const LEAPING_INTO_ACTION_PREFIX = "Leaping into action";
 
-function buildCommentFooter(payload: Payload): string {
+async function buildCommentFooter(payload: Payload, octokit?: Octokit): Promise<string> {
   const repoContext = parseRepoContext();
   const runId = process.env.GITHUB_RUN_ID;
 
   const agentName = payload.agent;
   const agentInfo = agentName ? agentsManifest[agentName] : null;
 
+  let workflowRunHtmlUrl: string | undefined;
+  if (runId && octokit) {
+    try {
+      // fetch jobs to get the job URL for deep linking
+      const { data: jobs } = await octokit.rest.actions.listJobsForWorkflowRun({
+        owner: repoContext.owner,
+        repo: repoContext.name,
+        run_id: parseInt(runId, 10),
+      });
+      // use the first job's URL if available
+      workflowRunHtmlUrl = jobs.jobs[0]?.html_url ?? undefined;
+    } catch {
+      // fall back to building URL from runId if jobs can't be fetched
+    }
+  }
+
   return buildPullfrogFooter({
     triggeredBy: true,
     agent: {
       displayName: agentInfo?.displayName || "Unknown agent",
       url: agentInfo?.url || "https://pullfrog.com",
     },
-    workflowRun: runId ? { owner: repoContext.owner, repo: repoContext.name, runId } : undefined,
+    workflowRun: runId
+      ? {
+          owner: repoContext.owner,
+          repo: repoContext.name,
+          runId,
+          ...(workflowRunHtmlUrl ? { htmlUrl: workflowRunHtmlUrl } : {}),
+        }
+      : undefined,
   });
 }
 
-function addFooter(body: string, payload: Payload): string {
+async function addFooter(body: string, payload: Payload, octokit?: Octokit): Promise<string> {
   const bodyWithoutFooter = stripExistingFooter(body);
-  const footer = buildCommentFooter(payload);
+  const footer = await buildCommentFooter(payload, octokit);
   return `${bodyWithoutFooter}${footer}`;
 }
 
@@ -43,14 +66,14 @@ export const Comment = type({
   body: type.string.describe("the comment body content"),
 });
 
-export function CreateCommentTool(ctx: Context) {
+export function CreateCommentTool(ctx: ToolContext) {
   return tool({
     name: "create_issue_comment",
     description:
       "Create a comment on a GitHub issue. NOTE: Do NOT use this for progress updates or status summaries - use report_progress instead, which updates the existing progress comment.",
     parameters: Comment,
-    execute: execute(ctx, async ({ issueNumber, body }) => {
-      const bodyWithFooter = addFooter(body, ctx.payload);
+    execute: execute(async ({ issueNumber, body }) => {
+      const bodyWithFooter = await addFooter(body, ctx.payload, ctx.octokit);
 
       const result = await ctx.octokit.rest.issues.createComment({
         owner: ctx.owner,
@@ -74,13 +97,13 @@ export const EditComment = type({
   body: type.string.describe("the new comment body content"),
 });
 
-export function EditCommentTool(ctx: Context) {
+export function EditCommentTool(ctx: ToolContext) {
   return tool({
     name: "edit_issue_comment",
     description: "Edit a GitHub issue comment by its ID",
     parameters: EditComment,
-    execute: execute(ctx, async ({ commentId, body }) => {
-      const bodyWithFooter = addFooter(body, ctx.payload);
+    execute: execute(async ({ commentId, body }) => {
+      const bodyWithFooter = await addFooter(body, ctx.payload, ctx.octokit);
 
       const result = await ctx.octokit.rest.issues.updateComment({
         owner: ctx.owner,
@@ -147,7 +170,7 @@ export const ReportProgress = type({
  * Returns result data if successful, undefined if comment cannot be created.
  */
 export async function reportProgress(
-  ctx: Context,
+  ctx: ToolContext,
   { body }: { body: string }
 ): Promise<
   | {
@@ -158,7 +181,7 @@ export async function reportProgress(
     }
   | undefined
 > {
-  const bodyWithFooter = addFooter(body, ctx.payload);
+  const bodyWithFooter = await addFooter(body, ctx.payload, ctx.octokit);
   const existingCommentId = getProgressCommentId();
 
   // if we already have a progress comment, update it
@@ -181,7 +204,9 @@ export async function reportProgress(
   }
 
   // no existing comment - create one
-  const issueNumber = ctx.payload.event.issue_number;
+  // use fallback chain: dynamically set context > event payload
+  const issueNumber =
+    ctx.toolState.prNumber ?? ctx.toolState.issueNumber ?? ctx.payload.event.issue_number;
   if (issueNumber === undefined) {
     // cannot create comment without issue_number (e.g., workflow_dispatch events)
     return undefined;
@@ -206,13 +231,13 @@ export async function reportProgress(
   };
 }
 
-export function ReportProgressTool(ctx: Context) {
+export function ReportProgressTool(ctx: ToolContext) {
   return tool({
     name: "report_progress",
     description:
       "Share progress on the associated GitHub issue/PR. Call this to post updates as you work. The first call creates a comment, subsequent calls update it. Use this throughout your work to keep stakeholders informed.",
     parameters: ReportProgress,
-    execute: execute(ctx, async ({ body }) => {
+    execute: execute(async ({ body }) => {
       const result = await reportProgress(ctx, { body });
 
       if (!result) {
@@ -244,7 +269,7 @@ export function wasProgressCommentUpdated(): boolean {
  * Delete the progress comment if it exists.
  * Used after submitting a PR review since the review body contains all necessary info.
  */
-export async function deleteProgressComment(ctx: Context): Promise<boolean> {
+export async function deleteProgressComment(ctx: ToolContext): Promise<boolean> {
   const existingCommentId = getProgressCommentId();
   if (!existingCommentId) {
     return false;
@@ -337,7 +362,7 @@ export async function ensureProgressCommentUpdated(payload?: Payload): Promise<v
 The workflow encountered an error before any progress could be reported. Please check the ${workflowRunLink} for details.`;
 
   // add footer if we have payload, otherwise use plain message
-  const body = payload ? addFooter(errorMessage, payload) : errorMessage;
+  const body = payload ? await addFooter(errorMessage, payload, octokit) : errorMessage;
 
   await octokit.rest.issues.updateComment({
     owner: repoContext.owner,
@@ -355,14 +380,14 @@ export const ReplyToReviewComment = type({
   ),
 });
 
-export function ReplyToReviewCommentTool(ctx: Context) {
+export function ReplyToReviewCommentTool(ctx: ToolContext) {
   return tool({
     name: "reply_to_review_comment",
     description:
       "Reply to a PR review comment thread. Call this for EACH comment you address. Keep replies extremely brief (1 sentence max).",
     parameters: ReplyToReviewComment,
-    execute: execute(ctx, async ({ pull_number, comment_id, body }) => {
-      const bodyWithFooter = addFooter(body, ctx.payload);
+    execute: execute(async ({ pull_number, comment_id, body }) => {
+      const bodyWithFooter = await addFooter(body, ctx.payload, ctx.octokit);
 
       const result = await ctx.octokit.rest.pulls.createReplyForReviewComment({
         owner: ctx.owner,

mcp/debug.ts

@@ -1,24 +1,23 @@
 import { type } from "arktype";
+import type { ToolContext } from "../main.ts";
 import { $ } from "../utils/shell.ts";
-import { handleToolError, handleToolSuccess, tool, type ToolResult } from "./shared.ts";
+import { execute, tool } from "./shared.ts";
 
 export const DebugShellCommand = type({});
 
-export const DebugShellCommandTool = tool({
-  name: "debug_shell_command",
-  description:
-    "debug tool: runs 'git status' and returns the output. use this to test shell command execution in the MCP server.",
-  parameters: DebugShellCommand,
-  execute: async (): Promise<ToolResult> => {
-    try {
+export function DebugShellCommandTool(_ctx: ToolContext) {
+  return tool({
+    name: "debug_shell_command",
+    description:
+      "debug tool: runs 'git status' and returns the output. use this to test shell command execution in the MCP server.",
+    parameters: DebugShellCommand,
+    execute: execute(async () => {
       const result = $("git", ["status"]);
-      return handleToolSuccess({
+      return {
         success: true,
         command: "git status",
         output: result.trim(),
-      });
-    } catch (error) {
-      return handleToolError(error);
-    }
-  },
-});
+      };
+    }),
+  });
+}

mcp/dependencies.ts

@@ -0,0 +1,180 @@
+import { type } from "arktype";
+import type { ToolContext } from "../main.ts";
+import type { PrepResult } from "../prep/index.ts";
+import { runPrepPhase } from "../prep/index.ts";
+import { execute, tool } from "./shared.ts";
+
+// empty schema for tools with no parameters
+const EmptyParams = type({});
+
+/**
+ * format prep results into agent-friendly message
+ */
+function formatPrepResults(results: PrepResult[]): string {
+  if (results.length === 0) {
+    return `No supported language detected in this repository (checked for package.json, requirements.txt, pyproject.toml, etc.).
+
+Inspect the repository structure to determine how dependencies should be installed, then use bash to install them.`;
+  }
+
+  const lines: string[] = [];
+
+  for (const result of results) {
+    if (result.language === "unknown") {
+      continue;
+    }
+
+    const langDisplay = result.language === "node" ? "Node.js" : "Python";
+
+    if (result.dependenciesInstalled) {
+      if (result.language === "node") {
+        lines.push(
+          `${langDisplay} dependencies installed successfully via ${result.packageManager}.`
+        );
+      } else if (result.language === "python") {
+        lines.push(
+          `${langDisplay} dependencies installed successfully via ${result.packageManager} (from ${result.configFile}).`
+        );
+      }
+    } else {
+      const errorMsg = result.issues.length > 0 ? result.issues.join("\n") : "unknown error";
+
+      if (result.language === "node") {
+        lines.push(`${langDisplay} dependency installation failed via ${result.packageManager}.
+
+Error:
+${errorMsg}
+
+Use bash or other tools at your disposal to diagnose and resolve the issue, then install dependencies manually.`);
+      } else if (result.language === "python") {
+        lines.push(`${langDisplay} dependency installation failed via ${result.packageManager} (from ${result.configFile}).
+
+Error:
+${errorMsg}
+
+Use bash or other tools at your disposal to diagnose and resolve the issue, then install dependencies manually.`);
+      }
+    }
+  }
+
+  if (lines.length === 0) {
+    return `No supported language detected in this repository (checked for package.json, requirements.txt, pyproject.toml, etc.).
+
+Inspect the repository structure to determine how dependencies should be installed, then use bash to install them.`;
+  }
+
+  return lines.join("\n\n");
+}
+
+/**
+ * start dependency installation in the background (non-blocking, idempotent)
+ */
+function startInstallation(ctx: ToolContext): void {
+  // already started or completed - do nothing
+  if (ctx.toolState.dependencyInstallation) {
+    return;
+  }
+
+  // initialize state and start installation
+  const promise = runPrepPhase();
+  ctx.toolState.dependencyInstallation = {
+    status: "in_progress",
+    promise,
+    results: undefined,
+  };
+
+  // when promise completes, update state
+  promise.then(
+    (results) => {
+      if (ctx.toolState.dependencyInstallation) {
+        const hasFailure = results.some((r) => !r.dependenciesInstalled && r.issues.length > 0);
+        ctx.toolState.dependencyInstallation.status = hasFailure ? "failed" : "completed";
+        ctx.toolState.dependencyInstallation.results = results;
+      }
+    },
+    () => {
+      if (ctx.toolState.dependencyInstallation) {
+        ctx.toolState.dependencyInstallation.status = "failed";
+      }
+    }
+  );
+}
+
+export function StartDependencyInstallationTool(ctx: ToolContext) {
+  return tool({
+    name: "start_dependency_installation",
+    description:
+      "Start installing project dependencies in the background. This is non-blocking and returns immediately. Call this early (right after branch checkout) if you anticipate needing to run tests, builds, or other commands that require dependencies. Idempotent - safe to call multiple times.",
+    parameters: EmptyParams,
+    execute: execute(async () => {
+      const state = ctx.toolState.dependencyInstallation;
+
+      // already completed
+      if (state?.status === "completed" || state?.status === "failed") {
+        return {
+          status: state.status,
+          message: `Dependency installation already completed.`,
+          summary: formatPrepResults(state.results || []),
+        };
+      }
+
+      // already in progress
+      if (state?.status === "in_progress") {
+        return {
+          status: "in_progress",
+          message:
+            "Dependency installation is already in progress. Call await_dependency_installation when you need to use them.",
+        };
+      }
+
+      // start installation
+      startInstallation(ctx);
+
+      return {
+        status: "started",
+        message:
+          "Dependency installation started in background. Continue with other tasks and call await_dependency_installation when you need to run tests, builds, or other commands that require dependencies.",
+      };
+    }),
+  });
+}
+
+export function AwaitDependencyInstallationTool(ctx: ToolContext) {
+  return tool({
+    name: "await_dependency_installation",
+    description:
+      "Wait for dependency installation to complete and get the results. If installation hasn't been started yet, this will start it automatically. Call this before running tests, builds, or other commands that require dependencies.",
+    parameters: EmptyParams,
+    execute: execute(async () => {
+      // auto-start if not started
+      if (!ctx.toolState.dependencyInstallation) {
+        startInstallation(ctx);
+      }
+
+      const state = ctx.toolState.dependencyInstallation;
+      if (!state) {
+        throw new Error("failed to initialize dependency installation state");
+      }
+
+      // if already completed, return cached results
+      if (state.status === "completed" || state.status === "failed") {
+        return {
+          status: state.status,
+          message: formatPrepResults(state.results || []),
+        };
+      }
+
+      // await the promise
+      if (!state.promise) {
+        throw new Error("dependency installation state is corrupted - no promise found");
+      }
+
+      const results = await state.promise;
+
+      return {
+        status: state.status,
+        message: formatPrepResults(results),
+      };
+    }),
+  });
+}

mcp/files.ts

@@ -1,6 +1,8 @@
+import { relative, resolve } from "node:path";
 import { type } from "arktype";
+import type { ToolContext } from "../main.ts";
 import { $ } from "../utils/shell.ts";
-import { handleToolError, handleToolSuccess, tool, type ToolResult } from "./shared.ts";
+import { execute, tool } from "./shared.ts";
 
 export const ListFiles = type({
   path: type.string
@@ -8,38 +10,82 @@ export const ListFiles = type({
     .default("."),
 });
 
-// static tool - doesn't need ctx, just runs git/find commands
-export const ListFilesTool = tool({
-  name: "list_files",
-  description:
-    "List files in the repository using git ls-files. Useful for discovering the file structure and locating files.",
-  parameters: ListFiles,
-  execute: async ({ path }: { path?: string }): Promise<ToolResult> => {
-    try {
-      // Use git ls-files to list tracked files
-      // This respects .gitignore and gives a clean list of source files
+export function ListFilesTool(_ctx: ToolContext) {
+  return tool({
+    name: "list_files",
+    description:
+      "List files in the repository, including both git-tracked and untracked files. Useful for discovering the file structure and locating files, including newly created files that haven't been committed yet.",
+    parameters: ListFiles,
+    execute: execute(async ({ path }: { path?: string }) => {
       const pathStr = path ?? ".";
-      const output = $("git", pathStr === "." ? ["ls-files"] : ["ls-files", pathStr], {
-        log: false,
-      });
-      const files = output.split("\n").filter((f) => f.trim() !== "");
+      const cwd = process.cwd();
 
-      if (files.length === 0) {
-        // Fallback for non-git environments or untracked files
+      // Get git-tracked files
+      let gitFiles: string[] = [];
+      let gitFailed = false;
+      try {
+        const gitArgs = pathStr === "." ? ["ls-files"] : ["ls-files", pathStr];
+        const gitOutput = $("git", gitArgs, { log: false });
+        gitFiles = gitOutput
+          .split("\n")
+          .filter((f) => f.trim() !== "")
+          .map((f) => f.trim());
+      } catch {
+        // git might fail, that's ok - we'll use find instead
+        gitFailed = true;
+      }
+
+      // Always also check filesystem for untracked files
+      // This is important because newly created files won't be in git yet
+      let filesystemFiles: string[] = [];
+      let findFailed = false;
+      try {
         const findOutput = $(
           "find",
           [pathStr, "-maxdepth", "3", "-not", "-path", "*/.*", "-type", "f"],
           { log: false }
         );
-        return handleToolSuccess({
-          files: findOutput.split("\n").filter((f) => f.trim() !== ""),
-          method: "find",
-        });
+        filesystemFiles = findOutput
+          .split("\n")
+          .filter((f) => f.trim() !== "")
+          .map((f) => {
+            const trimmed = f.trim();
+            // normalize to relative paths for comparison
+            try {
+              return relative(cwd, resolve(cwd, trimmed));
+            } catch {
+              return trimmed;
+            }
+          });
+      } catch {
+        // find might fail, that's ok - we'll just use git files
+        findFailed = true;
       }
 
-      return handleToolSuccess({ files, method: "git" });
-    } catch (error) {
-      return handleToolError(error);
-    }
-  },
-});
+      // if both methods failed, throw an error (execute helper will handle it)
+      if (gitFailed && findFailed) {
+        throw new Error(
+          `Failed to list files: both git ls-files and find commands failed. ` +
+            `Path: ${pathStr}, working directory: ${cwd}`
+        );
+      }
+
+      // Create a Set of git files for efficient lookup
+      const gitFilesSet = new Set(gitFiles);
+
+      // Combine both lists, removing duplicates
+      const allFiles = [...new Set([...gitFiles, ...filesystemFiles])].sort();
+
+      // Calculate actual untracked count (files in filesystem but not in git)
+      const untrackedFiles = filesystemFiles.filter((f) => !gitFilesSet.has(f));
+      const untrackedCount = untrackedFiles.length;
+
+      return {
+        files: allFiles,
+        method: "combined",
+        trackedCount: gitFiles.length,
+        untrackedCount,
+      };
+    }),
+  });
+}

mcp/git.ts

@@ -1,11 +1,11 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { log } from "../utils/cli.ts";
 import { containsSecrets } from "../utils/secrets.ts";
 import { $ } from "../utils/shell.ts";
 import { execute, tool } from "./shared.ts";
 
-export function CreateBranchTool(ctx: Context) {
+export function CreateBranchTool(ctx: ToolContext) {
   const defaultBranch = ctx.repo.default_branch || "main";
 
   const CreateBranch = type({
@@ -22,7 +22,7 @@ export function CreateBranchTool(ctx: Context) {
     description:
       "Create a new git branch from the specified base branch. The branch will be created locally and pushed to the remote repository.",
     parameters: CreateBranch,
-    execute: execute(ctx, async ({ branchName, baseBranch }) => {
+    execute: execute(async ({ branchName, baseBranch }) => {
       // baseBranch should always be defined due to default, but TypeScript needs help
       const resolvedBaseBranch = baseBranch || ctx.repo.default_branch || "main";
 
@@ -34,7 +34,7 @@ export function CreateBranchTool(ctx: Context) {
         );
       }
 
-      log.info(`Creating branch ${branchName} from ${resolvedBaseBranch}`);
+      log.debug(`Creating branch ${branchName} from ${resolvedBaseBranch}`);
 
       // fetch base branch to ensure we're up to date
       $("git", ["fetch", "origin", resolvedBaseBranch, "--depth=1"]);
@@ -49,7 +49,7 @@ export function CreateBranchTool(ctx: Context) {
       // push branch to remote (set upstream)
       $("git", ["push", "-u", "origin", branchName]);
 
-      log.info(`Successfully created and pushed branch ${branchName}`);
+      log.debug(`Successfully created and pushed branch ${branchName}`);
 
       return {
         success: true,
@@ -70,13 +70,13 @@ export const CommitFiles = type({
     ),
 });
 
-export function CommitFilesTool(ctx: Context) {
+export function CommitFilesTool(_ctx: ToolContext) {
   return tool({
     name: "commit_files",
     description:
       "Stage and commit files with a commit message. If files array is empty, commits all staged changes. The commit will be attributed to the correct bot account.",
     parameters: CommitFiles,
-    execute: execute(ctx, async ({ message, files }) => {
+    execute: execute(async ({ message, files }) => {
       // validate commit message for secrets
       if (containsSecrets(message)) {
         throw new Error(
@@ -109,7 +109,7 @@ export function CommitFilesTool(ctx: Context) {
       }
 
       const currentBranch = $("git", ["rev-parse", "--abbrev-ref", "HEAD"], { log: false });
-      log.info(`Committing files on branch ${currentBranch}`);
+      log.debug(`Committing files on branch ${currentBranch}`);
 
       // stage files if provided, otherwise stage all changes
       if (files.length > 0) {
@@ -122,7 +122,7 @@ export function CommitFilesTool(ctx: Context) {
       $("git", ["commit", "-m", message]);
 
       const commitSha = $("git", ["rev-parse", "HEAD"], { log: false });
-      log.info(`Successfully committed: ${commitSha.substring(0, 7)}`);
+      log.debug(`Successfully committed: ${commitSha.substring(0, 7)}`);
 
       return {
         success: true,
@@ -141,35 +141,52 @@ export const PushBranch = type({
   force: type.boolean.describe("Force push (use with caution)").default(false),
 });
 
-export function PushBranchTool(ctx: Context) {
-  const remote = ctx.pushRemote;
-
+export function PushBranchTool(_ctx: ToolContext) {
   return tool({
     name: "push_branch",
     description:
-      "Push the current branch (or specified branch) to the remote repository. Never force push unless explicitly requested.",
+      "Push the current branch (or specified branch) to the remote repository. Git automatically determines the correct remote based on branch config (set by checkout_pr for fork PRs). Never force push unless explicitly requested.",
     parameters: PushBranch,
-    execute: execute(ctx, async ({ branchName, force }) => {
+    execute: execute(async ({ branchName, force }) => {
       const branch = branchName || $("git", ["rev-parse", "--abbrev-ref", "HEAD"], { log: false });
 
-      // skip -u flag when pushing to URL (can't set upstream without a remote name)
-      const isUrl = remote.startsWith("https://");
-      const args = force
-        ? ["push", "--force", ...(isUrl ? [] : ["-u"]), remote, branch]
-        : ["push", ...(isUrl ? [] : ["-u"]), remote, branch];
+      // check if branch has a configured pushRemote
+      let remote = "origin";
+      try {
+        remote = $("git", ["config", `branch.${branch}.pushRemote`], { log: false }).trim();
+      } catch {
+        // no configured pushRemote, default to origin
+      }
 
-      log.info(`Pushing branch ${branch} to ${isUrl ? "(fork URL)" : remote}`);
+      // check if branch has a configured merge ref (remote branch name may differ from local)
+      let remoteBranch = branch;
+      try {
+        const mergeRef = $("git", ["config", `branch.${branch}.merge`], { log: false }).trim();
+        // merge ref is like "refs/heads/main", extract the branch name
+        remoteBranch = mergeRef.replace("refs/heads/", "");
+      } catch {
+        // no configured merge ref, use local branch name
+      }
+
+      // use refspec when local and remote branch names differ
+      const refspec = branch === remoteBranch ? branch : `${branch}:${remoteBranch}`;
+      const args = force
+        ? ["push", "--force", "-u", remote, refspec]
+        : ["push", "-u", remote, refspec];
+
+      log.debug(`pushing ${branch} to ${remote}/${remoteBranch}`);
       if (force) {
-        log.warning(`Force pushing - this will overwrite remote history`);
+        log.warning(`force pushing - this will overwrite remote history`);
       }
       $("git", args);
 
       return {
         success: true,
         branch,
-        remote: isUrl ? "(fork URL)" : remote,
+        remoteBranch,
+        remote,
         force,
-        message: `Successfully pushed branch ${branch}`,
+        message: `successfully pushed ${branch} to ${remote}/${remoteBranch}`,
       };
     }),
   });

mcp/issue.ts

@@ -1,5 +1,5 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const Issue = type({
@@ -15,12 +15,12 @@ export const Issue = type({
     .optional(),
 });
 
-export function IssueTool(ctx: Context) {
+export function IssueTool(ctx: ToolContext) {
   return tool({
     name: "create_issue",
     description: "Create a new GitHub issue",
     parameters: Issue,
-    execute: execute(ctx, async ({ title, body, labels, assignees }) => {
+    execute: execute(async ({ title, body, labels, assignees }) => {
       const result = await ctx.octokit.rest.issues.create({
         owner: ctx.owner,
         repo: ctx.name,
@@ -37,7 +37,9 @@ export function IssueTool(ctx: Context) {
         url: result.data.html_url,
         title: result.data.title,
         state: result.data.state,
-        labels: result.data.labels?.map((label) => (typeof label === "string" ? label : label.name)),
+        labels: result.data.labels?.map((label) =>
+          typeof label === "string" ? label : label.name
+        ),
         assignees: result.data.assignees?.map((assignee) => assignee.login),
       };
     }),

mcp/issueComments.ts

@@ -1,18 +1,21 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const GetIssueComments = type({
   issue_number: type.number.describe("The issue number to get comments for"),
 });
 
-export function GetIssueCommentsTool(ctx: Context) {
+export function GetIssueCommentsTool(ctx: ToolContext) {
   return tool({
     name: "get_issue_comments",
     description:
       "Get all comments for a GitHub issue. Returns all comments including the issue body and all subsequent discussion comments.",
     parameters: GetIssueComments,
-    execute: execute(ctx, async ({ issue_number }) => {
+    execute: execute(async ({ issue_number }) => {
+      // set issue context
+      ctx.toolState.issueNumber = issue_number;
+
       const comments = await ctx.octokit.paginate(ctx.octokit.rest.issues.listComments, {
         owner: ctx.owner,
         repo: ctx.name,

mcp/issueEvents.ts

@@ -1,18 +1,21 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const GetIssueEvents = type({
   issue_number: type.number.describe("The issue number to get events for"),
 });
 
-export function GetIssueEventsTool(ctx: Context) {
+export function GetIssueEventsTool(ctx: ToolContext) {
   return tool({
     name: "get_issue_events",
     description:
       "Get timeline events for a GitHub issue that aren't reflected in the current state. Returns cross-references to other issues/PRs and commit references. Note: current labels, assignees, state, and milestone are already available via get_issue.",
     parameters: GetIssueEvents,
-    execute: execute(ctx, async ({ issue_number }) => {
+    execute: execute(async ({ issue_number }) => {
+      // set issue context
+      ctx.toolState.issueNumber = issue_number;
+
       const events = await ctx.octokit.paginate(ctx.octokit.rest.issues.listEventsForTimeline, {
         owner: ctx.owner,
         repo: ctx.name,

mcp/issueInfo.ts

@@ -1,17 +1,17 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const IssueInfo = type({
   issue_number: type.number.describe("The issue number to fetch"),
 });
 
-export function IssueInfoTool(ctx: Context) {
+export function IssueInfoTool(ctx: ToolContext) {
   return tool({
     name: "get_issue",
     description: "Retrieve GitHub issue information by issue number",
     parameters: IssueInfo,
-    execute: execute(ctx, async ({ issue_number }) => {
+    execute: execute(async ({ issue_number }) => {
       const issue = await ctx.octokit.rest.issues.get({
         owner: ctx.owner,
         repo: ctx.name,
@@ -20,6 +20,9 @@ export function IssueInfoTool(ctx: Context) {
 
       const data = issue.data;
 
+      // set issue context
+      ctx.toolState.issueNumber = issue_number;
+
       const hints: string[] = [];
       if (data.comments > 0) {
         hints.push("use get_issue_comments to retrieve all comments for this issue");

mcp/labels.ts

@@ -1,5 +1,5 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const AddLabelsParams = type({
@@ -7,13 +7,13 @@ export const AddLabelsParams = type({
   labels: type.string.array().atLeastLength(1).describe("array of label names to add"),
 });
 
-export function AddLabelsTool(ctx: Context) {
+export function AddLabelsTool(ctx: ToolContext) {
   return tool({
     name: "add_labels",
     description:
       "Add labels to a GitHub issue or pull request. Only use labels that already exist in the repository.",
     parameters: AddLabelsParams,
-    execute: execute(ctx, async ({ issue_number, labels }) => {
+    execute: execute(async ({ issue_number, labels }) => {
       const result = await ctx.octokit.rest.issues.addLabels({
         owner: ctx.owner,
         repo: ctx.name,

mcp/pr.ts

@@ -1,5 +1,7 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import { agentsManifest } from "../external.ts";
+import type { ToolContext } from "../main.ts";
+import { buildPullfrogFooter, stripExistingFooter } from "../utils/buildPullfrogFooter.ts";
 import { log } from "../utils/cli.ts";
 import { containsSecrets } from "../utils/secrets.ts";
 import { $ } from "../utils/shell.ts";
@@ -11,14 +13,30 @@ export const PullRequest = type({
   base: type.string.describe("the base branch to merge into (e.g., 'main')"),
 });
 
-export function PullRequestTool(ctx: Context) {
+function buildPrBodyWithFooter(ctx: ToolContext, body: string): string {
+  const agentName = ctx.payload.agent;
+  const agentInfo = agentName ? agentsManifest[agentName] : null;
+
+  const footer = buildPullfrogFooter({
+    triggeredBy: true,
+    agent: agentInfo ? { displayName: agentInfo.displayName, url: agentInfo.url } : undefined,
+    workflowRun: ctx.runId
+      ? { owner: ctx.owner, repo: ctx.name, runId: ctx.runId, jobId: ctx.jobId }
+      : undefined,
+  });
+
+  const bodyWithoutFooter = stripExistingFooter(body);
+  return `${bodyWithoutFooter}${footer}`;
+}
+
+export function PullRequestTool(ctx: ToolContext) {
   return tool({
     name: "create_pull_request",
     description: "Create a pull request from the current branch",
     parameters: PullRequest,
-    execute: execute(ctx, async ({ title, body, base }) => {
+    execute: execute(async ({ title, body, base }) => {
       const currentBranch = $("git", ["rev-parse", "--abbrev-ref", "HEAD"], { log: false });
-      log.info(`Current branch: ${currentBranch}`);
+      log.debug(`Current branch: ${currentBranch}`);
 
       // validate PR title and body for secrets
       if (containsSecrets(title) || containsSecrets(body)) {
@@ -29,7 +47,9 @@ export function PullRequestTool(ctx: Context) {
       }
 
       // validate all changes that would be in the PR (from base to HEAD)
-      const diff = $("git", ["diff", `origin/${base}...HEAD`], { log: false });
+      // FORK PR NOTE: origin/<base> is fetched by setupGit, so this works for both fork and same-repo PRs
+      // use two-dot (..) not three-dot (...) for reliable diffs with shallow clones
+      const diff = $("git", ["diff", `origin/${base}..HEAD`], { log: false });
       if (containsSecrets(diff)) {
         throw new Error(
           "PR creation blocked: secrets detected in changes. " +
@@ -37,11 +57,13 @@ export function PullRequestTool(ctx: Context) {
         );
       }
 
+      const bodyWithFooter = buildPrBodyWithFooter(ctx, body);
+
       const result = await ctx.octokit.rest.pulls.create({
         owner: ctx.owner,
         repo: ctx.name,
         title: title,
-        body: body,
+        body: bodyWithFooter,
         head: currentBranch,
         base: base,
       });

mcp/prInfo.ts

@@ -1,18 +1,18 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const PullRequestInfo = type({
   pull_number: type.number.describe("The pull request number to fetch"),
 });
 
-export function PullRequestInfoTool(ctx: Context) {
+export function PullRequestInfoTool(ctx: ToolContext) {
   return tool({
     name: "get_pull_request",
     description:
-      "Retrieve PR information (metadata only). PR branch is already checked out during setup.",
+      "Retrieve PR metadata (number, title, state, base/head branches, fork status). To checkout a PR branch locally, use checkout_pr instead.",
     parameters: PullRequestInfo,
-    execute: execute(ctx, async ({ pull_number }) => {
+    execute: execute(async ({ pull_number }) => {
       const pr = await ctx.octokit.rest.pulls.get({
         owner: ctx.owner,
         repo: ctx.name,
@@ -21,8 +21,8 @@ export function PullRequestInfoTool(ctx: Context) {
 
       const data = pr.data;
 
-      // detect fork PRs - head repo differs from base repo
-      const isFork = data.head.repo.full_name !== data.base.repo.full_name;
+      // detect fork PRs - head repo differs from base repo (head.repo can be null if fork was deleted)
+      const isFork = data.head.repo?.full_name !== data.base.repo.full_name;
 
       return {
         number: data.number,

mcp/review.ts

@@ -1,10 +1,256 @@
 import type { RestEndpointMethodTypes } from "@octokit/rest";
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { buildPullfrogFooter } from "../utils/buildPullfrogFooter.ts";
+import { log } from "../utils/cli.ts";
 import { deleteProgressComment } from "./comment.ts";
 import { execute, tool } from "./shared.ts";
 
+// graphql mutation to add a comment thread to a pending review
+// note: REST API doesn't support adding comments to an existing pending review
+const ADD_PULL_REQUEST_REVIEW_THREAD = `
+mutation AddPullRequestReviewThread($pullRequestReviewId: ID!, $path: String!, $line: Int!, $body: String!, $side: DiffSide) {
+  addPullRequestReviewThread(input: {
+    pullRequestReviewId: $pullRequestReviewId,
+    path: $path,
+    line: $line,
+    body: $body,
+    side: $side
+  }) {
+    thread {
+      id
+    }
+  }
+}
+`;
+
+type AddPullRequestReviewThreadResponse = {
+  addPullRequestReviewThread: {
+    thread: {
+      id: string;
+    };
+  };
+};
+
+// helper to find existing pending review for the authenticated user
+async function findPendingReview(
+  ctx: ToolContext,
+  pull_number: number
+): Promise<{ id: number; node_id: string } | null> {
+  const reviews = await ctx.octokit.rest.pulls.listReviews({
+    owner: ctx.owner,
+    repo: ctx.name,
+    pull_number,
+    per_page: 100,
+  });
+
+  // find a PENDING review from our bot
+  // note: authenticated user is the GitHub App, reviews show as "pullfrog[bot]"
+  const pendingReview = reviews.data.find((r) => r.state === "PENDING");
+  if (pendingReview) {
+    return { id: pendingReview.id, node_id: pendingReview.node_id };
+  }
+  return null;
+}
+
+// start_review tool
+export const StartReview = type({
+  pull_number: type.number.describe("The pull request number to review"),
+});
+
+export function StartReviewTool(ctx: ToolContext) {
+  return tool({
+    name: "start_review",
+    description:
+      "Start a new review session for a pull request. Creates a pending review on GitHub. Must be called before add_review_comment.",
+    parameters: StartReview,
+    execute: execute(async ({ pull_number }) => {
+      // check if review already started in this session
+      if (ctx.toolState.review) {
+        throw new Error(
+          `Review session already in progress. Call submit_review first to finish it.`
+        );
+      }
+
+      // get the PR to get head commit SHA
+      const pr = await ctx.octokit.rest.pulls.get({
+        owner: ctx.owner,
+        repo: ctx.name,
+        pull_number,
+      });
+
+      let reviewId: number;
+      let reviewNodeId: string;
+
+      // try to create a new pending review (omitting 'event' creates PENDING state)
+      log.debug(`creating pending review for PR #${pull_number}...`);
+      try {
+        const result = await ctx.octokit.rest.pulls.createReview({
+          owner: ctx.owner,
+          repo: ctx.name,
+          pull_number,
+          commit_id: pr.data.head.sha,
+          // no 'event' = PENDING review
+        });
+        reviewId = result.data.id;
+        reviewNodeId = result.data.node_id;
+        log.debug(`created new pending review: id=${reviewId}`);
+      } catch (error) {
+        // check for "already has pending review" error
+        const errorMessage = error instanceof Error ? error.message : String(error);
+        log.debug(`createReview failed: ${errorMessage}`);
+        if (errorMessage.includes("pending review")) {
+          // find the existing pending review
+          log.debug(`pending review already exists, fetching existing review...`);
+          const existing = await findPendingReview(ctx, pull_number);
+          if (!existing) {
+            throw new Error(
+              "GitHub says a pending review exists but we couldn't find it. Try again or check the PR reviews."
+            );
+          }
+          reviewId = existing.id;
+          reviewNodeId = existing.node_id;
+          log.debug(`reusing existing pending review: id=${reviewId}`);
+        } else {
+          throw error;
+        }
+      }
+
+      // set PR context and review state
+      ctx.toolState.prNumber = pull_number;
+      ctx.toolState.review = {
+        nodeId: reviewNodeId,
+        id: reviewId,
+      };
+
+      return {
+        message: `Review session started for PR #${pull_number}.`,
+        guidance: {
+          analyze: [
+            "What does this PR change? Summarize in 1-2 sentences.",
+            "Is the approach sound? If not, stop and comment on approach first.",
+            "What bugs, edge cases, or security issues exist?",
+          ],
+          beforeCommenting: [
+            "Skip nitpicks unless explicitly requested.",
+            "Would the codebase maintainer care about this feedback, based on what you can infer about the code quality standards in this repo?",
+          ],
+        },
+      };
+    }),
+  });
+}
+
+// add_review_comment tool
+export const AddReviewComment = type({
+  path: type.string.describe("The file path to comment on (relative to repo root)"),
+  line: type.number.describe(
+    "The line number in the file (use line numbers from the diff - the NEW file line number)"
+  ),
+  body: type.string.describe("The comment text for this specific line"),
+  side: type
+    .enumerated("LEFT", "RIGHT")
+    .describe("Side of the diff: LEFT (old code) or RIGHT (new code). Defaults to RIGHT.")
+    .optional(),
+});
+
+export function AddReviewCommentTool(ctx: ToolContext) {
+  return tool({
+    name: "add_review_comment",
+    description:
+      "Add a comment to the current review session. Must call start_review first. Comments are stored in draft state until submit_review is called.",
+    parameters: AddReviewComment,
+    execute: execute(async ({ path, line, body, side }) => {
+      // check if review started
+      if (!ctx.toolState.review) {
+        throw new Error("No review session started. Call start_review first.");
+      }
+
+      // add comment thread via GraphQL (REST doesn't support adding to existing pending review)
+      await ctx.octokit.graphql<AddPullRequestReviewThreadResponse>(
+        ADD_PULL_REQUEST_REVIEW_THREAD,
+        {
+          pullRequestReviewId: ctx.toolState.review.nodeId,
+          path,
+          line,
+          body,
+          side: side || "RIGHT",
+        }
+      );
+
+      return {
+        success: true,
+        message: `Comment added to ${path}:${line}`,
+      };
+    }),
+  });
+}
+
+// submit_review tool
+export const SubmitReview = type({
+  body: type.string
+    .describe(
+      "Review body text. Typically 1-3 sentences with high-level overview and urgency level. Action links are auto-appended."
+    )
+    .optional(),
+});
+
+export function SubmitReviewTool(ctx: ToolContext) {
+  return tool({
+    name: "submit_review",
+    description:
+      "Submit the current review session. All comments added via add_review_comment will be published. Must call start_review first.",
+    parameters: SubmitReview,
+    execute: execute(async ({ body }) => {
+      // check if review started
+      if (!ctx.toolState.review) {
+        throw new Error("No review session started. Call start_review first.");
+      }
+      if (ctx.toolState.prNumber === undefined) {
+        throw new Error("No PR context. Call checkout_pr or start_review first.");
+      }
+
+      const reviewId = ctx.toolState.review.id;
+
+      // build quick links footer
+      const apiUrl = process.env.API_URL || "https://pullfrog.com";
+      const fixAllUrl = `${apiUrl}/trigger/${ctx.owner}/${ctx.name}/${ctx.toolState.prNumber}?action=fix&review_id=${reviewId}`;
+      const fixApprovedUrl = `${apiUrl}/trigger/${ctx.owner}/${ctx.name}/${ctx.toolState.prNumber}?action=fix-approved&review_id=${reviewId}`;
+
+      const footer = buildPullfrogFooter({
+        workflowRun: { owner: ctx.owner, repo: ctx.name, runId: ctx.runId, jobId: ctx.jobId },
+        customParts: [`[Fix all ➔](${fixAllUrl})`, `[Fix 👍s ➔](${fixApprovedUrl})`],
+      });
+
+      const bodyWithFooter = (body || "") + footer;
+
+      // submit the pending review via REST
+      const result = await ctx.octokit.rest.pulls.submitReview({
+        owner: ctx.owner,
+        repo: ctx.name,
+        pull_number: ctx.toolState.prNumber,
+        review_id: reviewId,
+        event: "COMMENT",
+        body: bodyWithFooter,
+      });
+
+      // clear review state
+      delete ctx.toolState.review;
+
+      // delete progress comment
+      await deleteProgressComment(ctx);
+
+      return {
+        success: true,
+        reviewId: result.data.id,
+        html_url: result.data.html_url,
+        state: result.data.state,
+      };
+    }),
+  });
+}
+
+// legacy tool - kept for backwards compatibility
 export const Review = type({
   pull_number: type.number.describe("The pull request number to review"),
   body: type.string
@@ -35,20 +281,25 @@ export const Review = type({
   })
     .array()
     .describe(
-      "PRIMARY location for ALL feedback. 95%+ of review content should be here. Use 'git diff origin/<base>...origin/<head>' to find correct line numbers (RIGHT side for new code, LEFT for old)."
+      // FORK PR NOTE: checkout_pr returns the diff via GitHub API - use that for line numbers
+      "PRIMARY location for ALL feedback. 95%+ of review content should be here. Use the diff returned from checkout_pr to find correct line numbers (RIGHT side for new code, LEFT for old)."
     )
     .optional(),
 });
 
-export function ReviewTool(ctx: Context) {
+export function ReviewTool(ctx: ToolContext) {
   return tool({
     name: "submit_pull_request_review",
     description:
+      "DEPRECATED: Use start_review, add_review_comment, and submit_review instead for iterative review workflow. " +
       "Submit a review for an existing pull request. " +
       "IMPORTANT: 95%+ of feedback should be in 'comments' array with file paths and line numbers. " +
       "Only use 'body' for a 1-2 sentence summary with urgency and critical callouts.",
     parameters: Review,
-    execute: execute(ctx, async ({ pull_number, body, commit_id, comments = [] }) => {
+    execute: execute(async ({ pull_number, body, commit_id, comments = [] }) => {
+      // set PR context
+      ctx.toolState.prNumber = pull_number;
+
       // get the PR to determine the head commit if commit_id not provided
       const pr = await ctx.octokit.rest.pulls.get({
         owner: ctx.owner,
@@ -93,6 +344,7 @@ export function ReviewTool(ctx: Context) {
       const fixApprovedUrl = `${apiUrl}/trigger/${ctx.owner}/${ctx.name}/${pull_number}?action=fix-approved&review_id=${reviewId}`;
 
       const footer = buildPullfrogFooter({
+        workflowRun: { owner: ctx.owner, repo: ctx.name, runId: ctx.runId, jobId: ctx.jobId },
         customParts: [`[Fix all ➔](${fixAllUrl})`, `[Fix 👍s ➔](${fixApprovedUrl})`],
       });
 

mcp/reviewComments.ts

@@ -1,5 +1,5 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 // graphql query to fetch all review threads with comments and replies
@@ -86,13 +86,13 @@ export const GetReviewComments = type({
   review_id: type.number.describe("The review ID to get comments for"),
 });
 
-export function GetReviewCommentsTool(ctx: Context) {
+export function GetReviewCommentsTool(ctx: ToolContext) {
   return tool({
     name: "get_review_comments",
     description:
       "Get all review comments and their replies for a specific pull request review. Returns line-by-line comments that were left on specific code locations, including any threaded replies.",
     parameters: GetReviewComments,
-    execute: execute(ctx, async ({ pull_number, review_id }) => {
+    execute: execute(async ({ pull_number, review_id }) => {
       // fetch all review threads using graphql
       const response = await ctx.octokit.graphql<GraphQLResponse>(REVIEW_THREADS_QUERY, {
         owner: ctx.owner,
@@ -189,13 +189,13 @@ export const ListPullRequestReviews = type({
   pull_number: type.number.describe("The pull request number to list reviews for"),
 });
 
-export function ListPullRequestReviewsTool(ctx: Context) {
+export function ListPullRequestReviewsTool(ctx: ToolContext) {
   return tool({
     name: "list_pull_request_reviews",
     description:
       "List all reviews for a pull request. Returns all reviews including approvals, request changes, and comments.",
     parameters: ListPullRequestReviews,
-    execute: execute(ctx, async ({ pull_number }) => {
+    execute: execute(async ({ pull_number }) => {
       const reviews = await ctx.octokit.paginate(ctx.octokit.rest.pulls.listReviews, {
         owner: ctx.owner,
         repo: ctx.name,

mcp/selectMode.ts

@@ -1,5 +1,5 @@
 import { type } from "arktype";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 import { execute, tool } from "./shared.ts";
 
 export const SelectMode = type({
@@ -8,13 +8,13 @@ export const SelectMode = type({
   ),
 });
 
-export function SelectModeTool(ctx: Context) {
+export function SelectModeTool(ctx: ToolContext) {
   return tool({
     name: "select_mode",
     description:
       "Select a mode and get its detailed prompt instructions. Call this first to determine which mode to use based on the request.",
     parameters: SelectMode,
-    execute: execute(ctx, async ({ modeName }) => {
+    execute: execute(async ({ modeName }) => {
       const selectedMode = ctx.modes.find((m) => m.name.toLowerCase() === modeName.toLowerCase());
 
       if (!selectedMode) {

mcp/server.ts

@@ -1,9 +1,10 @@
 import "./arkConfig.ts";
-// this must be imported first
 import { createServer } from "node:net";
+// this must be imported first
 import { FastMCP, type Tool } from "fastmcp";
 import { ghPullfrogMcpName } from "../external.ts";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
+import { CheckoutPrTool } from "./checkout.ts";
 import { GetCheckSuiteLogsTool } from "./checkSuite.ts";
 import {
   CreateCommentTool,
@@ -12,6 +13,11 @@ import {
   ReportProgressTool,
 } from "./comment.ts";
 import { DebugShellCommandTool } from "./debug.ts";
+import {
+  AwaitDependencyInstallationTool,
+  StartDependencyInstallationTool,
+} from "./dependencies.ts";
+import { ListFilesTool } from "./files.ts";
 import { CommitFilesTool, CreateBranchTool, PushBranchTool } from "./git.ts";
 import { IssueTool } from "./issue.ts";
 import { GetIssueCommentsTool } from "./issueComments.ts";
@@ -20,10 +26,10 @@ import { IssueInfoTool } from "./issueInfo.ts";
 import { AddLabelsTool } from "./labels.ts";
 import { PullRequestTool } from "./pr.ts";
 import { PullRequestInfoTool } from "./prInfo.ts";
-import { ReviewTool } from "./review.ts";
+import { AddReviewCommentTool, StartReviewTool, SubmitReviewTool } from "./review.ts";
 import { GetReviewCommentsTool, ListPullRequestReviewsTool } from "./reviewComments.ts";
 import { SelectModeTool } from "./selectMode.ts";
-import { addTools, isProgressCommentDisabled } from "./shared.ts";
+import { addTools } from "./shared.ts";
 
 /**
  * Find an available port starting from the given port
@@ -58,7 +64,7 @@ async function findAvailablePort(startPort: number): Promise<number> {
  * Start the MCP HTTP server and return the URL and close function
  */
 export async function startMcpHttpServer(
-  ctx: Context
+  ctx: ToolContext
 ): Promise<{ url: string; close: () => Promise<void> }> {
   const server = new FastMCP({
     name: ghPullfrogMcpName,
@@ -68,6 +74,8 @@ export async function startMcpHttpServer(
   // create all tools as factories, passing ctx
   const tools: Tool<any, any>[] = [
     SelectModeTool(ctx),
+    StartDependencyInstallationTool(ctx),
+    AwaitDependencyInstallationTool(ctx),
     CreateCommentTool(ctx),
     EditCommentTool(ctx),
     ReplyToReviewCommentTool(ctx),
@@ -76,20 +84,24 @@ export async function startMcpHttpServer(
     GetIssueCommentsTool(ctx),
     GetIssueEventsTool(ctx),
     PullRequestTool(ctx),
-    ReviewTool(ctx),
+    // ReviewTool(ctx),
+    StartReviewTool(ctx),
+    AddReviewCommentTool(ctx),
+    SubmitReviewTool(ctx),
     PullRequestInfoTool(ctx),
+    CheckoutPrTool(ctx),
     GetReviewCommentsTool(ctx),
     ListPullRequestReviewsTool(ctx),
     GetCheckSuiteLogsTool(ctx),
-    DebugShellCommandTool,
+    DebugShellCommandTool(ctx),
     AddLabelsTool(ctx),
     CreateBranchTool(ctx),
     CommitFilesTool(ctx),
     PushBranchTool(ctx),
+    ListFilesTool(ctx),
   ];
 
-  // only include ReportProgressTool if progress comment is not disabled
-  if (!isProgressCommentDisabled(ctx)) {
+  if (!ctx.payload.disableProgressComment) {
     tools.push(ReportProgressTool(ctx));
   }
 

mcp/shared.ts

@@ -1,6 +1,6 @@
 import type { StandardSchemaV1 } from "@standard-schema/spec";
 import type { FastMCP, Tool } from "fastmcp";
-import type { Context } from "../main.ts";
+import type { ToolContext } from "../main.ts";
 
 export const tool = <const params>(toolDef: Tool<any, StandardSchemaV1<params>>) => toolDef;
 
@@ -40,7 +40,7 @@ export const handleToolError = (error: unknown): ToolResult => {
  * Helper to wrap a tool execute function with error handling.
  * Captures ctx in closure so tools don't need to handle try/catch.
  */
-export const execute = <T>(ctx: Context, fn: (params: T) => Promise<Record<string, any>>) => {
+export const execute = <T>(fn: (params: T) => Promise<Record<string, any>>) => {
   return async (params: T): Promise<ToolResult> => {
     try {
       const result = await fn(params);
@@ -51,10 +51,6 @@ export const execute = <T>(ctx: Context, fn: (params: T) => Promise<Record<strin
   };
 };
 
-export function isProgressCommentDisabled(ctx: Context): boolean {
-  return ctx.payload.disableProgressComment === true;
-}
-
 /**
  * Sanitize JSON schema to remove problematic fields that Gemini CLI/API can't handle
  * - Removes $schema field (causes "no schema with key or ref" errors)
@@ -176,10 +172,10 @@ function sanitizeTool<T extends Tool<any, any>>(tool: T): T {
   } as T;
 }
 
-export const addTools = (ctx: Context, server: FastMCP, tools: Tool<any, any>[]) => {
+export const addTools = (ctx: ToolContext, server: FastMCP, tools: Tool<any, any>[]) => {
   // sanitize schemas for gemini agent and opencode (when using Google API)
   // both have issues with draft-2020-12 schemas and any_of enum constructs
-  const shouldSanitize = ctx.agentName === "gemini" || ctx.agentName === "opencode";
+  const shouldSanitize = ctx.agent.name === "gemini" || ctx.agent.name === "opencode";
 
   for (const tool of tools) {
     const processedTool = shouldSanitize ? sanitizeTool(tool) : tool;

modes.ts

@@ -12,6 +12,17 @@ export interface GetModesParams {
 
 const reportProgressInstruction = `Use ${ghPullfrogMcpName}/report_progress to share progress and results. Continue calling it as you make progress - it will update the same comment. Never create additional comments manually.`;
 
+const dependencyInstallationGuidance = `## Dependency Installation
+
+**IMPORTANT**: Immediately after the working branch is checked out, evaluate whether dependencies will be needed at any point during this task:
+- Making code changes that will require testing? → Call \`${ghPullfrogMcpName}/start_dependency_installation\` NOW
+- Running builds, linters, or CLI commands that require installed packages? → Call \`${ghPullfrogMcpName}/start_dependency_installation\` NOW
+- Only reading code or answering questions? → Skip dependency installation
+
+Calling \`start_dependency_installation\` early allows dependencies to install in the background while you explore the codebase and make changes. This is a non-blocking call.
+
+When you need to run tests, builds, or other commands that require dependencies, call \`${ghPullfrogMcpName}/await_dependency_installation\` to ensure they're ready. This will block until installation completes (or auto-start if you forgot to call start earlier).`;
+
 export function getModes({ disableProgressComment }: GetModesParams): Mode[] {
   return [
     {
@@ -19,9 +30,11 @@ export function getModes({ disableProgressComment }: GetModesParams): Mode[] {
       description:
         "Implement, build, create, or develop code changes; make specific changes to files or features; execute a plan; or handle tasks with specific implementation details",
       prompt: `Follow these steps:
-1. If the request requires understanding the codebase structure, dependencies, or conventions, gather relevant context. Read AGENTS.md if it exists, understand how to install dependencies, run tests, run builds, and make changes according to best practices). Skip this step if the prompt is trivial and self-contained.
+1. If this is a PR event, the PR branch is already checked out - skip branch creation. Otherwise, create a branch using ${ghPullfrogMcpName}/create_branch. The branch name should be prefixed with "pullfrog/". The rest of the name should reflect the exact changes you are making. It should be specific to avoid collisions with other branches. Never commit directly to main, master, or production. Do NOT use git commands directly (including \`git branch\`, \`git status\`, \`git log\`) - always use ${ghPullfrogMcpName} MCP tools for git operations.
 
-2. Create a branch using ${ghPullfrogMcpName}/create_branch. The branch name should be prefixed with "pullfrog/". The rest of the name should reflect the exact changes you are making. It should be specific to avoid collisions with other branches. Never commit directly to main, master, or production. Do NOT use git commands directly - always use ${ghPullfrogMcpName} MCP tools for git operations.
+${dependencyInstallationGuidance}
+
+2. If the request requires understanding the codebase structure or conventions, gather relevant context. Read AGENTS.md if it exists. Skip this step if the prompt is trivial and self-contained.
 
 3. Understand the requirements and any existing plan
 
@@ -56,11 +69,15 @@ export function getModes({ disableProgressComment }: GetModesParams): Mode[] {
       description:
         "Address PR review feedback; respond to reviewer comments; make requested changes to an existing PR",
       prompt: `Follow these steps:
-1. Get PR info with ${ghPullfrogMcpName}/get_pull_request (this automatically fetches and checks out the PR branch)
+1. Checkout the PR using ${ghPullfrogMcpName}/checkout_pr with the PR number. This fetches the PR branch and configures push settings (including for fork PRs).
+
+${dependencyInstallationGuidance}
 
 2. Review the feedback provided. Understand each review comment and what changes are being requested.
+   - **EVENT DATA may contain review comment details**: If available, \`approved_comments\` are comments to address, \`unapproved_comments\` are for context only. The \`triggerer\` field indicates who initiated this action - prioritize their replies when deciding how to implement fixes.
+   - You can use ${ghPullfrogMcpName}/get_pull_request to get PR metadata if needed.
 
-3. If the request requires understanding the codebase structure, dependencies, or conventions, gather relevant context. Read AGENTS.md if it exists.
+3. If the request requires understanding the codebase structure or conventions, gather relevant context. Read AGENTS.md if it exists.
 
 4. Make the necessary code changes to address the feedback. Work through each review comment systematically.
 
@@ -68,7 +85,7 @@ export function getModes({ disableProgressComment }: GetModesParams): Mode[] {
 
 6. Test your changes to ensure they work correctly.
 
-7. When done, commit and push your changes to the existing PR branch. Do not create a new branch or PR - you are updating an existing one.
+7. When done, commit your changes with ${ghPullfrogMcpName}/commit_files, then push with ${ghPullfrogMcpName}/push_branch. The push will automatically go to the correct remote (including fork repos). Do not create a new branch or PR - you are updating an existing one.
 ${
   disableProgressComment
     ? ""
@@ -83,32 +100,35 @@ ${
       description:
         "Review code, PRs, or implementations; provide feedback or suggestions; identify issues; or check code quality, style, and correctness",
       prompt: `Follow these steps:
-1. Get PR info with ${ghPullfrogMcpName}/get_pull_request (this automatically prepares the repository by fetching and checking out the PR branch)
+1. Checkout the PR using ${ghPullfrogMcpName}/checkout_pr with the PR number. This fetches the PR branch and returns the PR diff in the \`diff\` field of the response. Use this diff for your review - it shows exactly what's in the PR (fetched via GitHub API, so it's not affected by main advancing after the branch was created).
 
-2. **IMPORTANT**: After calling ${ghPullfrogMcpName}/get_pull_request, the PR branch is already checked out locally. View diff using: \`git diff origin/<base>...HEAD\` (replace <base> with 'base' from PR info). Do NOT use \`origin/<head>\` - the branch is checked out locally, not as a remote tracking branch.
+2. Start review session using ${ghPullfrogMcpName}/start_review. This creates a pending review on GitHub and returns analysis guidance. Follow the guidance before adding comments.
 
-3. Read files from the checked-out PR branch to understand the implementation. Always use **relative paths** from repo root (e.g., \`src/index.ts\`), never absolute paths.
+3. **ANALYZE** - Before adding any comments, think through:
+   - What does this PR change? Summarize in 1-2 sentences.
+   - Is the approach sound? If not, **stop here** and comment on the approach first. Don't waste time on implementation details if the approach is wrong.
+   - What bugs, edge cases, or security issues exist?
 
-4. Submit review using ${ghPullfrogMcpName}/submit_pull_request_review
+4. **BEFORE COMMENTING** - For each potential comment, ask yourself:
+   - Is this a nitpick? Skip it unless explicitly requested.
+   - Would the codebase maintainer care about this feedback, based on what you can infer about the code quality standards in this repo?
+
+5. Add inline review comments one-by-one using ${ghPullfrogMcpName}/add_review_comment
+   - Use **relative paths** from repo root (e.g., \`packages/core/src/utils.ts\`)
+   - Use the NEW file line number from the diff (shown after \`+\` in hunk headers like \`@@ -10,5 +12,8 @@\` means new file starts at line 12)
+   - Only comment on lines that appear in the diff. GitHub will reject comments on unchanged lines.
+   - For issues appearing in multiple places, comment on the FIRST occurrence and reference others (e.g., "also at lines X, Y")
+
+6. Submit the review using ${ghPullfrogMcpName}/submit_review
+   - The "body" field is ONLY for: (1) a 1-3 sentence high-level overview, (2) urgency level (e.g., "minor suggestions" vs "blocking issues"), (3) critical security callouts (e.g., API key exposure)
 
 **GENERAL GUIDANCE**
 
-- *CRITICAL* — Use **relative paths** from repo root (e.g., \`packages/core/src/utils.ts\`)
-  - For line numbers, use the NEW file line number from the diff (shown after \`+\` in hunk headers like \`@@ -10,5 +12,8 @@\` means new file starts at line 12)
-  - Only comment on lines that appear in the diff. GitHub will reject comments on unchanged lines.
+- Do not leave any comments that are not potentially actionable. Do not leave complimentary comments just to be nice.
+- Do not nitpick unless instructed explicitly to do so by the user's additional instructions. This includes: requesting documentation/docstrings/JSDoc.
 - **CRITICAL: Prioritize per-line feedback over summary text.**
-  - ALL specific feedback MUST go in the 'comments' array with file paths and line numbers from the diff
-  - For issues appearing in multiple places, comment on the FIRST occurrence and reference others (e.g., "also at lines X, Y" or "similar issue in otherFile.ts:42")
-  - The "body" field is ONLY for: (1) a 1-2 sentence high-level overview, (2) urgency level (e.g., "minor suggestions" vs "blocking issues"), (3) critical security callouts (e.g., API key exposure)
-  - 95%+ of review content should be in per-line comments; the body should be just a couple sentences
-  - The review body will include quick action links for addressing feedback, so keep it concise
-- Do not nitpick unless instructed explicity to do so by the user's additional instructions. This includes: requesting documentation/docstrings/JSDoc. 
-- Do not leave any comments that are not potentially actionable. 
-- The review should be thoughtful. When evaluating complex changes, consider the following conceptual approach:
-  - 1) conceptualize the changes made. make sure you understand it.
-  - 2) evaluate conceptual approach. leave feedback as needed.
-  - 3) if the conceptual approach looks sound, evaluate the implementation. leave feedback as needed. consider everything, but especially edge cases, security, correctness, and performance. leave feedback as needed.
-  - 4) only leave nitpick/housekeeping comments if instructed explicity to do so by the user's additional instructions.
+  - All specific feedback MUST go in inline review comments with file paths and line numbers from the diff
+  - The vast majority of review content should be in inline review comments; the body should be brief and only summarize the urgency of the review and any cross-cutting concerns.
   `,
     },
     {
@@ -116,7 +136,7 @@ ${
       description:
         "Create plans, break down tasks, outline steps, analyze requirements, understand scope of work, or provide task breakdowns",
       prompt: `Follow these steps:
-1. If the request requires understanding the codebase structure, dependencies, or conventions, gather relevant context (read AGENTS.md if it exists, understand how to install dependencies, run tests, run builds, and make changes according to best practices). Skip this step if the prompt is trivial and self-contained.
+1. If the request requires understanding the codebase structure or conventions, gather relevant context (read AGENTS.md if it exists). Skip this step if the prompt is trivial and self-contained.
 
 2. Analyze the request and break it down into clear, actionable tasks
 
@@ -133,6 +153,9 @@ ${
 
 2. If the task involves making code changes:
    - Create a branch using ${ghPullfrogMcpName}/create_branch. Branch names should be prefixed with "pullfrog/" and reflect the exact changes you are making. Never commit directly to main, master, or production.
+
+${dependencyInstallationGuidance}
+
    - Use file operations to create/modify files with your changes.
    - Use ${ghPullfrogMcpName}/commit_files to commit your changes, then ${ghPullfrogMcpName}/push_branch to push the branch. Do NOT use git commands directly (\`git commit\`, \`git push\`, \`git checkout\`, \`git branch\`) as these will use incorrect credentials.
    - Test your changes to ensure they work correctly.
@@ -145,4 +168,6 @@ ${
   ];
 }
 
-export const modes: Mode[] = getModes({ disableProgressComment: undefined });
+export const modes: Mode[] = getModes({
+  disableProgressComment: undefined,
+});

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pullfrog/action",
-  "version": "0.0.144",
+  "version": "0.0.149",
   "type": "module",
   "files": [
     "index.js",
@@ -34,6 +34,7 @@
     "@opencode-ai/sdk": "^1.0.143",
     "@standard-schema/spec": "1.0.0",
     "arktype": "2.1.28",
+    "package-manager-detector": "^1.6.0",
     "dotenv": "^17.2.3",
     "execa": "^9.6.0",
     "fastmcp": "^3.20.0",

play.ts

@@ -19,7 +19,7 @@ config({ path: join(process.cwd(), "..", ".env") });
 export async function run(prompt: string): Promise<AgentResult> {
   try {
     const tempDir = join(process.cwd(), ".temp");
-    setupTestRepo({ tempDir, forceClean: true });
+    setupTestRepo({ tempDir });
 
     const originalCwd = process.cwd();
     process.chdir(tempDir);

pnpm-lock.yaml

@@ -50,6 +50,9 @@ importers:
       fastmcp:
         specifier: ^3.20.0
         version: 3.20.0(arktype@2.1.28)
+      package-manager-detector:
+        specifier: ^1.6.0
+        version: 1.6.0
       table:
         specifier: ^6.9.0
         version: 6.9.0
@@ -836,6 +839,9 @@ packages:
   once@1.4.0:
     resolution: {integrity: sha512-lNaJgI+2Q5URQBkccEKHTQOPaXdUxnZZElQTZY0MFUAuaEqe1E+Nyvgdz/aIyNi6Z9MzO5dv1H8n58/GELp3+w==}
 
+  package-manager-detector@1.6.0:
+    resolution: {integrity: sha512-61A5ThoTiDG/C8s8UMZwSorAGwMJ0ERVGj2OjoW5pAalsNOg15+iQiPzrLJ4jhZ1HJzmC2PIHT2oEiH3R5fzNA==}
+
   parse-ms@4.0.0:
     resolution: {integrity: sha512-TXfryirbmq34y8QBwgqCVLi+8oA3oWx2eAnSn62ITyEhEYaWRlVZ2DvMM9eZbMs/RfxPu/PK/aBLyGj4IrqMHw==}
     engines: {node: '>=18'}
@@ -1843,6 +1849,8 @@ snapshots:
     dependencies:
       wrappy: 1.0.2
 
+  package-manager-detector@1.6.0: {}
+
   parse-ms@4.0.0: {}
 
   parseurl@1.3.3: {}

prep/index.ts

@@ -0,0 +1,42 @@
+import { log } from "../utils/cli.ts";
+import { installNodeDependencies } from "./installNodeDependencies.ts";
+import { installPythonDependencies } from "./installPythonDependencies.ts";
+import type { PrepDefinition, PrepResult } from "./types.ts";
+
+export type { PrepResult } from "./types.ts";
+
+// register all prep steps here
+const prepSteps: PrepDefinition[] = [installNodeDependencies, installPythonDependencies];
+
+/**
+ * run all prep steps sequentially.
+ * failures are logged as warnings but don't stop the run.
+ */
+export async function runPrepPhase(): Promise<PrepResult[]> {
+  log.debug("» starting prep phase...");
+  const startTime = Date.now();
+  const results: PrepResult[] = [];
+
+  for (const step of prepSteps) {
+    const shouldRun = await step.shouldRun();
+    if (!shouldRun) {
+      log.debug(`» skipping ${step.name} (not applicable)`);
+      continue;
+    }
+
+    log.debug(`» running ${step.name}...`);
+    const result = await step.run();
+    results.push(result);
+
+    if (result.dependenciesInstalled) {
+      log.debug(`» ${step.name}: dependencies installed`);
+    } else if (result.issues.length > 0) {
+      log.warning(`⚠️  ${step.name}: ${result.issues[0]}`);
+    }
+  }
+
+  const totalDurationMs = Date.now() - startTime;
+  log.debug(`» prep phase completed (${totalDurationMs}ms)`);
+
+  return results;
+}

prep/installNodeDependencies.ts

@@ -0,0 +1,160 @@
+import { existsSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+import { isKeyOf } from "@ark/util";
+import { detect } from "package-manager-detector";
+import { resolveCommand } from "package-manager-detector/commands";
+import { log } from "../utils/cli.ts";
+import { spawn } from "../utils/subprocess.ts";
+import type { NodePackageManager, NodePrepResult, PrepDefinition } from "./types.ts";
+
+// install command templates for each package manager (version placeholder: {version})
+const nodePackageManagers: Record<NodePackageManager, string[]> = {
+  npm: ["echo", "npm is already installed"],
+  pnpm: ["npm", "install", "-g", "{version}"],
+  yarn: ["npm", "install", "-g", "{version}"],
+  bun: ["npm", "install", "-g", "{version}"],
+  deno: ["sh", "-c", "curl -fsSL https://deno.land/install.sh | sh"],
+};
+
+async function isCommandAvailable(command: string): Promise<boolean> {
+  const result = await spawn({
+    cmd: "which",
+    args: [command],
+    env: { PATH: process.env.PATH || "" },
+  });
+  return result.exitCode === 0;
+}
+
+interface PackageManagerSpec {
+  name: NodePackageManager;
+  installSpec: string; // e.g., "pnpm@8.15.0" (without hash suffix)
+}
+
+function getPackageManagerFromPackageJson(): PackageManagerSpec | null {
+  const packageJsonPath = join(process.cwd(), "package.json");
+  try {
+    const content = readFileSync(packageJsonPath, "utf-8");
+    const pkg = JSON.parse(content) as { packageManager?: string };
+    if (!pkg.packageManager) return null;
+
+    // format: "pnpm@8.15.0" or "pnpm@8.15.0+sha512.abc123..."
+    // strip the hash suffix (+sha256.xxx) as npm install doesn't understand it
+    const withoutHash = pkg.packageManager.split("+")[0];
+    const name = withoutHash.split("@")[0];
+    if (isKeyOf(name, nodePackageManagers)) {
+      return { name, installSpec: withoutHash };
+    }
+    log.warning(`unknown packageManager in package.json: ${pkg.packageManager}`);
+    return null;
+  } catch {
+    return null;
+  }
+}
+
+async function installPackageManager(
+  name: NodePackageManager,
+  installSpec: string
+): Promise<string | null> {
+  if (name === "npm") return null; // npm is always available
+  log.info(`📦 installing ${installSpec}...`);
+  const [cmd, ...templateArgs] = nodePackageManagers[name];
+  const args = templateArgs.map((arg) => (arg === "{version}" ? installSpec : arg));
+  const result = await spawn({
+    cmd,
+    args,
+    env: { PATH: process.env.PATH || "", HOME: process.env.HOME || "" },
+    onStderr: (chunk) => process.stderr.write(chunk),
+  });
+
+  if (result.exitCode !== 0) {
+    return result.stderr || `failed to install ${name}`;
+  }
+
+  // deno installs to $HOME/.deno/bin - add to PATH for subsequent commands
+  if (name === "deno") {
+    const denoPath = join(process.env.HOME || "", ".deno", "bin");
+    process.env.PATH = `${denoPath}:${process.env.PATH}`;
+  }
+
+  log.info(`✅ installed ${name}`);
+  return null;
+}
+
+export const installNodeDependencies: PrepDefinition = {
+  name: "installNodeDependencies",
+
+  shouldRun: () => {
+    const packageJsonPath = join(process.cwd(), "package.json");
+    return existsSync(packageJsonPath);
+  },
+
+  run: async (): Promise<NodePrepResult> => {
+    // check packageManager field in package.json first (takes priority)
+    const fromPackageJson = getPackageManagerFromPackageJson();
+
+    // detect from lockfile as fallback
+    const detected = await detect({ cwd: process.cwd() });
+
+    // prefer package.json field, fall back to lockfile detection, default to npm
+    const packageManager = fromPackageJson?.name || (detected?.name as NodePackageManager) || "npm";
+    const installSpec = fromPackageJson?.installSpec || packageManager;
+    const agent = detected?.agent || packageManager;
+
+    if (fromPackageJson) {
+      log.info(`📦 using packageManager from package.json: ${fromPackageJson.installSpec}`);
+    } else if (detected) {
+      log.info(`📦 detected package manager: ${packageManager} (${agent})`);
+    } else {
+      log.info(`📦 no package manager detected, defaulting to npm`);
+    }
+
+    // check if package manager is available, install if needed
+    if (!(await isCommandAvailable(packageManager))) {
+      log.info(`${packageManager} not found, attempting to install...`);
+      const installError = await installPackageManager(packageManager, installSpec);
+      if (installError) {
+        return {
+          language: "node",
+          packageManager,
+          dependenciesInstalled: false,
+          issues: [installError],
+        };
+      }
+    }
+
+    // get the frozen install command (or fallback to regular install)
+    const resolved = resolveCommand(agent, "frozen", []) || resolveCommand(agent, "install", []);
+    if (!resolved) {
+      return {
+        language: "node",
+        packageManager,
+        dependenciesInstalled: false,
+        issues: [`no install command found for ${agent}`],
+      };
+    }
+
+    log.info(`running: ${resolved.command} ${resolved.args.join(" ")}`);
+    const result = await spawn({
+      cmd: resolved.command,
+      args: resolved.args,
+      env: { PATH: process.env.PATH || "", HOME: process.env.HOME || "" },
+      onStderr: (chunk) => process.stderr.write(chunk),
+    });
+
+    if (result.exitCode !== 0) {
+      return {
+        language: "node",
+        packageManager,
+        dependenciesInstalled: false,
+        issues: [result.stderr || `${resolved.command} exited with code ${result.exitCode}`],
+      };
+    }
+
+    return {
+      language: "node",
+      packageManager,
+      dependenciesInstalled: true,
+      issues: [],
+    };
+  },
+};

prep/installPythonDependencies.ts

@@ -0,0 +1,162 @@
+import { existsSync } from "node:fs";
+import { join } from "node:path";
+import { log } from "../utils/cli.ts";
+import { spawn } from "../utils/subprocess.ts";
+import type { PrepDefinition, PythonPackageManager, PythonPrepResult } from "./types.ts";
+
+interface PythonConfig {
+  file: string;
+  tool: PythonPackageManager;
+  installCmd: string[];
+}
+
+// python dependency file patterns in priority order
+const PYTHON_CONFIGS: PythonConfig[] = [
+  {
+    file: "requirements.txt",
+    tool: "pip",
+    installCmd: ["pip", "install", "-r", "requirements.txt"],
+  },
+  {
+    file: "pyproject.toml",
+    tool: "pip",
+    installCmd: ["pip", "install", "."],
+  },
+  {
+    file: "Pipfile",
+    tool: "pipenv",
+    installCmd: ["pipenv", "install"],
+  },
+  {
+    file: "Pipfile.lock",
+    tool: "pipenv",
+    installCmd: ["pipenv", "sync"],
+  },
+  {
+    file: "poetry.lock",
+    tool: "poetry",
+    installCmd: ["poetry", "install", "--no-interaction"],
+  },
+  {
+    file: "setup.py",
+    tool: "pip",
+    installCmd: ["pip", "install", "-e", "."],
+  },
+];
+
+// tool install commands (via pip)
+const TOOL_INSTALL_COMMANDS: Record<string, string[]> = {
+  pipenv: ["pip", "install", "pipenv"],
+  poetry: ["pip", "install", "poetry"],
+};
+
+async function isCommandAvailable(command: string): Promise<boolean> {
+  const result = await spawn({
+    cmd: "which",
+    args: [command],
+    env: { PATH: process.env.PATH || "" },
+  });
+  return result.exitCode === 0;
+}
+
+async function installTool(name: string): Promise<string | null> {
+  const installCmd = TOOL_INSTALL_COMMANDS[name];
+  if (!installCmd) {
+    // tool doesn't need installation (e.g., pip)
+    return null;
+  }
+
+  log.info(`📦 installing ${name}...`);
+  const [cmd, ...args] = installCmd;
+  const result = await spawn({
+    cmd,
+    args,
+    env: { PATH: process.env.PATH || "", HOME: process.env.HOME || "" },
+    onStderr: (chunk) => process.stderr.write(chunk),
+  });
+
+  if (result.exitCode !== 0) {
+    return result.stderr || `failed to install ${name}`;
+  }
+
+  log.info(`✅ installed ${name}`);
+  return null;
+}
+
+export const installPythonDependencies: PrepDefinition = {
+  name: "installPythonDependencies",
+
+  shouldRun: async () => {
+    // check if python is available
+    const hasPython = (await isCommandAvailable("python3")) || (await isCommandAvailable("python"));
+    if (!hasPython) {
+      return false;
+    }
+
+    // check if any python config file exists
+    const cwd = process.cwd();
+    return PYTHON_CONFIGS.some((config) => existsSync(join(cwd, config.file)));
+  },
+
+  run: async (): Promise<PythonPrepResult> => {
+    const cwd = process.cwd();
+
+    // find the first matching config
+    const config = PYTHON_CONFIGS.find((c) => existsSync(join(cwd, c.file)));
+    if (!config) {
+      return {
+        language: "python",
+        packageManager: "pip",
+        configFile: "unknown",
+        dependenciesInstalled: false,
+        issues: ["no python config file found"],
+      };
+    }
+
+    log.info(`🐍 detected python config: ${config.file} (using ${config.tool})`);
+
+    // check if the tool is available, install if needed
+    const isAvailable = await isCommandAvailable(config.tool);
+    if (!isAvailable) {
+      log.info(`${config.tool} not found, attempting to install...`);
+      const installError = await installTool(config.tool);
+      if (installError) {
+        return {
+          language: "python",
+          packageManager: config.tool,
+          configFile: config.file,
+          dependenciesInstalled: false,
+          issues: [installError],
+        };
+      }
+    }
+
+    // run the install command
+    const [cmd, ...args] = config.installCmd;
+    log.info(`running: ${cmd} ${args.join(" ")}`);
+    const result = await spawn({
+      cmd,
+      args,
+      env: { PATH: process.env.PATH || "", HOME: process.env.HOME || "" },
+      onStderr: (chunk) => process.stderr.write(chunk),
+    });
+
+    if (result.exitCode !== 0) {
+      return {
+        language: "python",
+        packageManager: config.tool,
+        configFile: config.file,
+        dependenciesInstalled: false,
+        issues: [result.stderr || `${cmd} exited with code ${result.exitCode}`],
+      };
+    }
+
+    return {
+      language: "python",
+      packageManager: config.tool,
+      configFile: config.file,
+      dependenciesInstalled: true,
+      issues: [],
+    };
+  },
+};

prep/types.ts

@@ -0,0 +1,31 @@
+interface PrepResultBase {
+  dependenciesInstalled: boolean;
+  issues: string[];
+}
+
+export type NodePackageManager = "npm" | "pnpm" | "yarn" | "bun" | "deno";
+
+export interface NodePrepResult extends PrepResultBase {
+  language: "node";
+  packageManager: NodePackageManager;
+}
+
+export type PythonPackageManager = "pip" | "pipenv" | "poetry";
+
+export interface PythonPrepResult extends PrepResultBase {
+  language: "python";
+  packageManager: PythonPackageManager;
+  configFile: string;
+}
+
+export interface UnknownLanguagePrepResult extends PrepResultBase {
+  language: "unknown";
+}
+
+export type PrepResult = NodePrepResult | PythonPrepResult | UnknownLanguagePrepResult;
+
+export interface PrepDefinition {
+  name: string;
+  shouldRun: () => Promise<boolean> | boolean;
+  run: () => Promise<PrepResult>;
+}

utils/buildPullfrogFooter.ts

@@ -7,10 +7,12 @@ export interface AgentInfo {
   url: string;
 }
 
-export interface WorkflowRunInfo {
+export interface WorkflowRunFooterInfo {
   owner: string;
   repo: string;
   runId: string;
+  /** optional job ID - if provided, will append /job/{jobId} to the workflow run URL */
+  jobId?: string | undefined;
 }
 
 export interface BuildPullfrogFooterParams {
@@ -19,7 +21,7 @@ export interface BuildPullfrogFooterParams {
   /** add "Using [agent](url)" link */
   agent?: AgentInfo | undefined;
   /** add "View workflow run" link */
-  workflowRun?: WorkflowRunInfo | undefined;
+  workflowRun?: WorkflowRunFooterInfo | undefined;
   /** arbitrary custom parts (e.g., action links) */
   customParts?: string[];
 }
@@ -39,16 +41,16 @@ export function buildPullfrogFooter(params: BuildPullfrogFooterParams): string {
     parts.push(`Using [${params.agent.displayName}](${params.agent.url})`);
   }
 
-  if (params.workflowRun) {
-    parts.push(
-      `[View workflow run](https://github.com/${params.workflowRun.owner}/${params.workflowRun.repo}/actions/runs/${params.workflowRun.runId})`
-    );
-  }
-
   if (params.customParts) {
     parts.push(...params.customParts);
   }
 
+  if (params.workflowRun) {
+    const baseUrl = `https://github.com/${params.workflowRun.owner}/${params.workflowRun.repo}/actions/runs/${params.workflowRun.runId}`;
+    const url = params.workflowRun.jobId ? `${baseUrl}/job/${params.workflowRun.jobId}` : baseUrl;
+    parts.push(`[View workflow run](${url})`);
+  }
+
   const allParts = [
     ...parts,
     "[pullfrog.com](https://pullfrog.com)",

utils/cli.ts

@@ -8,7 +8,11 @@ import * as core from "@actions/core";
 import { table } from "table";
 
 const isGitHubActions = !!process.env.GITHUB_ACTIONS;
-const isDebugEnabled = () => process.env.LOG_LEVEL === "debug";
+const isDebugEnabled = () =>
+  process.env.LOG_LEVEL === "debug" ||
+  process.env.ACTIONS_STEP_DEBUG === "true" ||
+  process.env.RUNNER_DEBUG === "1" ||
+  core.isDebug();
 
 /**
  * Start a collapsed group (GitHub Actions) or regular group (local)
@@ -32,6 +36,15 @@ function endGroup(): void {
   }
 }
 
+/**
+ * Run a callback within a collapsed group
+ */
+function group(name: string, fn: () => void): void {
+  startGroup(name);
+  fn();
+  endGroup();
+}
+
 /**
  * Print a formatted box with text (for console output)
  */
@@ -268,7 +281,10 @@ export const log = {
   debug: (message: string): void => {
     if (isDebugEnabled()) {
       if (isGitHubActions) {
-        core.debug(message);
+        // using this instead of core.debug
+        // because core.debug only logs when ACTIONS_STEP_DEBUG is set to true
+        // we are using LOG_LEVEL
+        core.info(`[DEBUG] ${message}`);
       } else {
         core.info(`[DEBUG] ${message}`);
       }
@@ -316,16 +332,22 @@ export const log = {
    */
   endGroup,
 
+  /**
+   * Run a callback within a collapsed group
+   */
+  group,
+
   /**
    * Log tool call information to console with formatted output
    */
   toolCall: ({ toolName, input }: { toolName: string; input: unknown }): void => {
-    let output = `→ ${toolName}\n`;
-
     const inputFormatted = formatJsonValue(input);
-    if (inputFormatted !== "{}") {
-      output += formatIndentedField("input", inputFormatted);
-    }
+    // if (inputFormatted !== "{}")
+    const output = inputFormatted !== "{}" ? `${toolName}(${inputFormatted})` : `${toolName}()`;
+
+    // if (inputFormatted !== "{}") {
+    // output += formatIndentedField("input", inputFormatted);
+    // }
 
     log.info(output.trimEnd());
   },

utils/github.ts

@@ -48,14 +48,14 @@ function isGitHubActionsEnvironment(): boolean {
 }
 
 async function acquireTokenViaOIDC(): Promise<string> {
-  log.info("Generating OIDC token...");
+  log.debug("» generating OIDC token...");
 
   const oidcToken = await core.getIDToken("pullfrog-api");
-  log.info("OIDC token generated successfully");
+  log.debug("» OIDC token generated successfully");
 
   const apiUrl = process.env.API_URL || "https://pullfrog.com";
 
-  log.info("Exchanging OIDC token for installation token...");
+  log.debug("» exchanging OIDC token for installation token...");
 
   // Add timeout to prevent long waits (5 seconds)
   const timeoutMs = 5000;
@@ -79,7 +79,7 @@ async function acquireTokenViaOIDC(): Promise<string> {
     }
 
     const tokenData = (await tokenResponse.json()) as InstallationToken;
-    log.info(`Installation token obtained for ${tokenData.repository || "all repositories"}`);
+    log.debug(`» installation token obtained for ${tokenData.repository || "all repositories"}`);
 
     return tokenData.token;
   } catch (error) {
@@ -283,7 +283,7 @@ export async function revokeGitHubInstallationToken(): Promise<void> {
         "X-GitHub-Api-Version": "2022-11-28",
       },
     });
-    log.info("Installation token revoked");
+    log.debug("» installation token revoked");
   } catch (error) {
     log.warning(
       `Failed to revoke installation token: ${error instanceof Error ? error.message : String(error)}`

utils/setup.ts

@@ -1,38 +1,29 @@
 import { execSync } from "node:child_process";
 import { existsSync, rmSync } from "node:fs";
-import type { Context } from "../main.ts";
+import type { Octokit } from "@octokit/rest";
+import type { Payload } from "../external.ts";
+import type { ToolState } from "../main.ts";
+import { checkoutPrBranch } from "../mcp/checkout.ts";
 import { log } from "./cli.ts";
 import { $ } from "./shell.ts";
 
 export interface SetupOptions {
   tempDir: string;
-  forceClean?: boolean;
 }
 
 /**
  * Setup the test repository for running actions
  */
 export function setupTestRepo(options: SetupOptions): void {
-  const { tempDir, forceClean = false } = options;
-
+  const { tempDir } = options;
+  const repo = process.env.GITHUB_REPOSITORY;
+  if (!repo) throw new Error("GITHUB_REPOSITORY is required");
   if (existsSync(tempDir)) {
-    if (forceClean) {
-      log.info("🗑️  Removing existing .temp directory...");
-      rmSync(tempDir, { recursive: true, force: true });
-
-      log.info("📦 Cloning pullfrog/scratch into .temp...");
-      $("git", ["clone", "git@github.com:pullfrog/scratch.git", tempDir]);
-    } else {
-      log.info("📦 Resetting existing .temp repository...");
-      execSync("git reset --hard HEAD && git clean -fd", {
-        cwd: tempDir,
-        stdio: "inherit",
-      });
-    }
-  } else {
-    log.info("📦 Cloning pullfrog/scratch into .temp...");
-    $("git", ["clone", "git@github.com:pullfrog/scratch.git", tempDir]);
+    log.info("» removing existing .temp directory...");
+    rmSync(tempDir, { recursive: true, force: true });
   }
+  log.info(`» cloning ${repo} into .temp...`);
+  $("git", ["clone", `git@github.com:${repo}.git`, tempDir]);
 }
 
 /**
@@ -41,7 +32,7 @@ export function setupTestRepo(options: SetupOptions): void {
  */
 export function setupGitConfig(): void {
   const repoDir = process.cwd();
-  log.info("🔧 Setting up git configuration...");
+  log.info("» setting up git configuration...");
   try {
     // Use --local to scope config to this repo only, preventing leakage to user's global config
     execSync('git config --local user.email "team@pullfrog.com"', {
@@ -52,7 +43,15 @@ export function setupGitConfig(): void {
       cwd: repoDir,
       stdio: "pipe",
     });
-    log.debug("setupGitConfig: ✓ Git configuration set successfully (scoped to repo)");
+    // disable credential helper to prevent macOS keychain prompts when using x-access-token
+    // only needed locally - GitHub Actions doesn't have this issue
+    if (!process.env.GITHUB_ACTIONS) {
+      execSync('git config --local credential.helper ""', {
+        cwd: repoDir,
+        stdio: "pipe",
+      });
+    }
+    log.debug("» git configuration set successfully (scoped to repo)");
   } catch (error) {
     // If git config fails, log warning but don't fail the action
     // This can happen if we're not in a git repo or git isn't available
@@ -62,19 +61,28 @@ export function setupGitConfig(): void {
   }
 }
 
-export type SetupGitResult = {
-  pushRemote: string;
-};
+interface SetupGitAuthParams {
+  token: string;
+  owner: string;
+  name: string;
+  payload: Payload;
+  octokit: Octokit;
+  toolState: ToolState;
+}
 
 /**
- * Unified git setup: configures authentication and checks out PR branch if applicable.
- * For PR events, always returns a push URL constructed from the PR's head repo.
- * This works for both same-repo and fork PRs with a single code path.
+ * Setup git authentication for the repository.
+ * For PR events, uses the shared checkoutPrBranch helper (also used by checkout_pr MCP tool).
+ *
+ * FORK PR ARCHITECTURE:
+ * - origin: always points to BASE REPO (where PR targets)
+ * - checkoutPrBranch sets per-branch pushRemote config for fork PRs
+ * - checkout_pr returns the PR diff via GitHub API (authoritative source)
  */
-export async function setupGit(ctx: Context): Promise<SetupGitResult> {
+export async function setupGitAuth(params: SetupGitAuthParams): Promise<void> {
   const repoDir = process.cwd();
 
-  log.info("🔧 Setting up git authentication...");
+  log.info("» setting up git authentication...");
 
   // remove existing git auth headers that actions/checkout might have set
   try {
@@ -82,45 +90,35 @@ export async function setupGit(ctx: Context): Promise<SetupGitResult> {
       cwd: repoDir,
       stdio: "pipe",
     });
-    log.info("✓ Removed existing authentication headers");
+    log.info("» removed existing authentication headers");
   } catch {
-    log.debug("No existing authentication headers to remove");
+    log.debug("» no existing authentication headers to remove");
   }
 
   // non-PR events: set up origin with token, stay on default branch
-  if (ctx.payload.event.is_pr !== true || !ctx.payload.event.issue_number) {
-    const originUrl = `https://x-access-token:${ctx.githubInstallationToken}@github.com/${ctx.owner}/${ctx.name}.git`;
+  if (params.payload.event.is_pr !== true || !params.payload.event.issue_number) {
+    const originUrl = `https://x-access-token:${params.token}@github.com/${params.owner}/${params.name}.git`;
     $("git", ["remote", "set-url", "origin", originUrl], { cwd: repoDir });
-    log.info("✓ Updated origin URL with authentication token");
-    return { pushRemote: "origin" };
+    log.info("» updated origin URL with authentication token");
+    return;
   }
 
-  // PR event: fetch PR info to get branch name and head repo
-  const prNumber = ctx.payload.event.issue_number;
-  const pr = await ctx.octokit.rest.pulls.get({
-    owner: ctx.owner,
-    repo: ctx.name,
-    pull_number: prNumber,
+  // PR event: checkout PR branch using shared helper
+  const prNumber = params.payload.event.issue_number;
+
+  // ensure origin is configured with auth token before checkout
+  const originUrl = `https://x-access-token:${params.token}@github.com/${params.owner}/${params.name}.git`;
+  $("git", ["remote", "set-url", "origin", originUrl], { cwd: repoDir });
+
+  // use shared checkout helper (handles fork remotes, push config, etc.)
+  const prContext = await checkoutPrBranch({
+    octokit: params.octokit,
+    owner: params.owner,
+    name: params.name,
+    token: params.token,
+    pullNumber: prNumber,
   });
 
-  const headRepo = pr.data.head.repo;
-  if (!headRepo) {
-    throw new Error(`PR #${prNumber} source repository was deleted`);
-  }
-
-  // checkout PR branch using plain git (no gh cli needed)
-  const branch = pr.data.head.ref;
-  log.info(`🌿 Checking out PR #${prNumber} (${branch})...`);
-  $("git", ["fetch", "origin", `pull/${prNumber}/head:${branch}`], { cwd: repoDir });
-  $("git", ["checkout", branch], { cwd: repoDir });
-  log.info(`✓ Successfully checked out PR #${prNumber}`);
-
-  // unified push URL - works for both same-repo and fork PRs
-  const token = process.env.GITHUB_TOKEN || ctx.githubInstallationToken;
-  const pushUrl = `https://x-access-token:${token}@github.com/${headRepo.full_name}.git`;
-  const isFork = headRepo.full_name !== pr.data.base.repo.full_name;
-  if (isFork) {
-    log.info(`🍴 Fork PR detected, will push to: ${headRepo.full_name}`);
-  }
-  return { pushRemote: pushUrl };
+  // set prNumber on toolState (the only mutation)
+  params.toolState.prNumber = prContext.prNumber;
 }

utils/timer.ts

@@ -14,7 +14,7 @@ export class Timer {
       ? now - this.lastCheckpointTimestamp
       : now - this.initialTimestamp;
 
-    log.info(`${name}: ${duration}ms`);
+    log.debug(`» ${name}: ${duration}ms`);
     this.lastCheckpointTimestamp = now;
   }
 }