0.0.157

add effort as an input + support parsing from payload

.github/workflows/test.yml

@@ -0,0 +1,35 @@
+name: Tests
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+          cache: "pnpm"
+
+      - name: Install dependencies
+        run: pnpm install --frozen-lockfile
+
+      - name: Run type check
+        run: pnpm typecheck
+
+      - name: Run tests
+        run: pnpm test

.husky/pre-commit

@@ -4,10 +4,3 @@ if git diff --cached --name-only | grep -q "^package.json$"; then
     pnpm lock
     git add pnpm-lock.yaml
 fi
-
-# Check if entry needs rebuilding (entry.ts, esbuild.config.js, or any .ts files)
-if git diff --cached --name-only | grep -qE "^(entry\.ts|esbuild\.config\.js|.*\.ts)$"; then
-    echo "🔨 Building action..."
-    pnpm build
-    git add entry
-fi

README.md

@@ -1,5 +1,4 @@
-<p align="center">
-  
+<p align="center">  
   <h1 align="center">
     <picture>
       <source media="(prefers-color-scheme: dark)" srcset="https://pullfrog.com/frog-white-200px.png">
@@ -84,6 +83,7 @@ on:
       prompt:
         description: 'Agent prompt'
         type: string
+    secrets: inherit
 
 permissions:
   id-token: write
@@ -105,14 +105,14 @@ jobs:
       - name: Run agent
         uses: pullfrog/action@v0
         with:
-          prompt: ${{ github.event.inputs.prompt }}
-
+          prompt: ${{ inputs.prompt }}
+        env:
           # feel free to comment out any you won't use
-          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
-          openai_api_key: ${{ secrets.OPENAI_API_KEY }}
-          google_api_key: ${{ secrets.GOOGLE_API_KEY }}
-          gemini_api_key: ${{ secrets.GEMINI_API_KEY }}
-          cursor_api_key: ${{ secrets.CURSOR_API_KEY }}
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          GOOGLE_API_KEY: ${{ secrets.GOOGLE_API_KEY }}
+          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
+          CURSOR_API_KEY: ${{ secrets.CURSOR_API_KEY }}
           
 ```
 

action.yml

@@ -1,30 +1,21 @@
-name: "Pullfrog Claude Code Action"
-description: "Execute Claude Code with a prompt using Anthropic API"
+name: "Pullfrog Action"
+description: "Execute coding agents with a prompt"
 author: "Pullfrog"
 
 inputs:
   prompt:
-    description: "Prompt to send to Claude Code"
+    description: "Prompt to send to the agent"
     required: true
-    default: "Hello from Claude Code!"
-  anthropic_api_key:
-    description: "Anthropic API key for Claude Code authentication"
+  effort:
+    description: "Effort level: nothink (fast), think (default), max (most capable)"
     required: false
-  openai_api_key:
-    description: "OpenAI API key for Codex authentication"
-    required: false
-  google_api_key:
-    description: "Google API key for Jules authentication"
-    required: false
-  gemini_api_key:
-    description: "Gemini API key for Jules authentication"
-    required: false
-  cursor_api_key:
-    description: "Cursor API key for Cursor authentication"
+    default: "think"
+  cwd:
+    description: "Working directory for the agent (defaults to GITHUB_WORKSPACE)"
     required: false
 
 runs:
-  using: "node20"
+  using: "node24"
   main: "entry"
 
 branding:

agents/claude.ts

@@ -1,9 +1,24 @@
 import { type Options, query, type SDKMessage } from "@anthropic-ai/claude-agent-sdk";
+import type { Effort } from "../external.ts";
 import packageJson from "../package.json" with { type: "json" };
 import { log } from "../utils/cli.ts";
 import { addInstructions } from "./instructions.ts";
 import { agent, createAgentEnv, installFromNpmTarball } from "./shared.ts";
 
+// Model selection based on effort level
+// Note: nothink uses Haiku for speed, think uses Sonnet for balance, max uses Opus for capability
+const claudeEffortModels: Record<Effort, string> = {
+  nothink: "haiku",
+  think: "opusplan",
+  max: "opus",
+};
+
+// FUTURE: Consider using Anthropic's "effort" parameter (beta) with Opus 4.5 for all tasks.
+// This would allow a single model with effort levels ("low", "medium", "high") controlling
+// token spend across responses, tool calls, and thinking. Requires beta header "effort-2025-11-24".
+// See: https://platform.claude.com/docs/en/build-with-claude/effort
+// This approach could replace model selection if effort proves effective for controlling capability.
+
 export const claude = agent({
   name: "claude",
   install: async () => {
@@ -14,13 +29,17 @@ export const claude = agent({
       executablePath: "cli.js",
     });
   },
-  run: async ({ payload, mcpServers, apiKey, cliPath, repo }) => {
+  run: async ({ payload, mcpServers, apiKey, cliPath, repo, effort }) => {
     // 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, repo });
     log.group("Full prompt", () => log.info(prompt));
 
+    // select model based on effort level
+    const model = claudeEffortModels[effort];
+    log.info(`Using model: ${model} (effort: ${effort})`);
+
     // SECURITY: For PUBLIC repos, Claude Code spawns subprocesses with full process.env, leaking API keys.
     // disable native Bash; agents use MCP bash tool which filters secrets.
     // for private repos, native Bash is allowed since secrets are less exposed.
@@ -46,15 +65,17 @@ export const claude = agent({
 
     // Pass secrets via SDK's env option only (not process.env)
     // This ensures secrets are only available to Claude Code subprocess, not user code
+    const queryOptions: Options = {
+      ...sandboxOptions,
+      mcpServers,
+      model,
+      pathToClaudeCodeExecutable: cliPath,
+      env: createAgentEnv({ ANTHROPIC_API_KEY: apiKey }),
+    };
+
     const queryInstance = query({
       prompt,
-      options: {
-        ...sandboxOptions,
-        mcpServers,
-        // model: "claude-opus-4-5",
-        pathToClaudeCodeExecutable: cliPath,
-        env: createAgentEnv({ ANTHROPIC_API_KEY: apiKey }),
-      },
+      options: queryOptions,
     });
 
     // Stream the results

agents/codex.ts

@@ -1,11 +1,33 @@
 import { mkdirSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 import type { McpHttpServerConfig } from "@anthropic-ai/claude-agent-sdk";
-import { Codex, type CodexOptions, type ThreadEvent } from "@openai/codex-sdk";
+import {
+  Codex,
+  type CodexOptions,
+  type ModelReasoningEffort,
+  type ThreadEvent,
+  type ThreadOptions,
+} from "@openai/codex-sdk";
+import type { Effort } from "../external.ts";
 import { log } from "../utils/cli.ts";
 import { addInstructions } from "./instructions.ts";
 import { agent, installFromNpmTarball, setupProcessAgentEnv } from "./shared.ts";
 
+// model configuration based on effort level
+const codexModel: Record<Effort, string> = {
+  nothink: "gpt-5.1-codex-mini",
+  think: "gpt-5.1-codex",
+  max: "gpt-5.1-codex-max",
+} as const;
+
+// reasoning effort configuration based on effort level
+// uses modelReasoningEffort parameter from ThreadOptions
+const codexReasoningEffort: Record<Effort, ModelReasoningEffort | undefined> = {
+  nothink: "low",
+  think: undefined, // use default
+  max: "high",
+};
+
 interface WriteCodexConfigParams {
   tempHome: string;
   mcpServers: Record<string, McpHttpServerConfig>;
@@ -60,7 +82,7 @@ export const codex = agent({
       executablePath: "bin/codex.js",
     });
   },
-  run: async ({ payload, mcpServers, apiKey, cliPath, repo }) => {
+  run: async ({ payload, mcpServers, apiKey, cliPath, repo, effort }) => {
     const tempHome = process.env.PULLFROG_TEMP_DIR!;
 
     // create config directory for codex before setting HOME
@@ -79,6 +101,14 @@ export const codex = agent({
       CODEX_HOME: codexDir, // point Codex to our config directory
     });
 
+    // get model and reasoning effort based on effort level
+    const model = codexModel[effort];
+    const modelReasoningEffort = codexReasoningEffort[effort];
+    log.info(`Using model: ${model}`);
+    if (modelReasoningEffort) {
+      log.info(`Using modelReasoningEffort: ${modelReasoningEffort}`);
+    }
+
     // Configure Codex
     const codexOptions: CodexOptions = {
       apiKey,
@@ -90,20 +120,28 @@ export const codex = agent({
     }
 
     const codex = new Codex(codexOptions);
-    const thread = codex.startThread(
-      payload.sandbox
-        ? {
-            approvalPolicy: "never",
-            sandboxMode: "read-only",
-            networkAccessEnabled: false,
-          }
-        : {
-            approvalPolicy: "never",
-            // use danger-full-access to allow git operations (workspace-write blocks .git directory writes)
-            sandboxMode: "danger-full-access",
-            networkAccessEnabled: true,
-          }
-    );
+
+    // Build thread options with model and optional model_reasoning_effort
+    const baseThreadOptions = payload.sandbox
+      ? {
+          model,
+          approvalPolicy: "never" as const,
+          sandboxMode: "read-only" as const,
+          networkAccessEnabled: false,
+        }
+      : {
+          model,
+          approvalPolicy: "never" as const,
+          // use danger-full-access to allow git operations (workspace-write blocks .git directory writes)
+          sandboxMode: "danger-full-access" as const,
+          networkAccessEnabled: true,
+        };
+
+    const threadOptions: ThreadOptions = modelReasoningEffort
+      ? { ...baseThreadOptions, modelReasoningEffort }
+      : baseThreadOptions;
+
+    const thread = codex.startThread(threadOptions);
 
     try {
       const streamedTurn = await thread.runStreamed(addInstructions({ payload, repo }));

agents/cursor.ts

@@ -1,7 +1,8 @@
 import { spawn } from "node:child_process";
-import { mkdirSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";
 import { join } from "node:path";
+import type { Effort } from "../external.ts";
 import { log } from "../utils/cli.ts";
 import { addInstructions } from "./instructions.ts";
 import {
@@ -11,6 +12,14 @@ import {
   installFromCurl,
 } from "./shared.ts";
 
+// effort configuration for Cursor
+// only "max" overrides the model; nothink/think use default ("auto")
+const cursorEffortModels: Record<Effort, string | null> = {
+  nothink: null, // use default (auto)
+  think: null, // use default (auto)
+  max: "opus-4.5-thinking",
+} as const;
+
 // cursor cli event types inferred from stream-json output
 interface CursorSystemEvent {
   type: "system";
@@ -91,10 +100,36 @@ export const cursor = agent({
       executableName: "cursor-agent",
     });
   },
-  run: async ({ payload, apiKey, cliPath, mcpServers, repo }) => {
+  run: async ({ payload, apiKey, cliPath, mcpServers, repo, effort }) => {
     configureCursorMcpServers({ mcpServers, cliPath });
     configureCursorSandbox({ sandbox: payload.sandbox ?? false, isPublicRepo: repo.isPublic });
 
+    // determine model based on effort level
+    // respect project's .cursor/cli.json if it specifies a model
+    const projectCliConfigPath = join(process.cwd(), ".cursor", "cli.json");
+    let modelOverride: string | null = null;
+
+    if (existsSync(projectCliConfigPath)) {
+      try {
+        const projectConfig = JSON.parse(readFileSync(projectCliConfigPath, "utf-8"));
+        if (projectConfig.model) {
+          log.info(`Using model from project .cursor/cli.json: ${projectConfig.model}`);
+        } else {
+          modelOverride = cursorEffortModels[effort];
+        }
+      } catch {
+        modelOverride = cursorEffortModels[effort];
+      }
+    } else {
+      modelOverride = cursorEffortModels[effort];
+    }
+
+    if (modelOverride) {
+      log.info(`Using model: ${modelOverride} (effort: ${effort})`);
+    } else if (!existsSync(projectCliConfigPath)) {
+      log.info(`Using default model (effort: ${effort})`);
+    }
+
     // track logged model_call_ids to avoid duplicates
     // cursor emits each assistant message twice: once without model_call_id, then again with it
     const loggedModelCallIds = new Set<string>();
@@ -171,16 +206,16 @@ export const cursor = agent({
 
       // configure sandbox mode if enabled
       // in sandbox mode: remove --force flag and rely on cli-config.json sandbox settings
+      const baseArgs = ["--print", fullPrompt, "--output-format", "stream-json", "--approve-mcps"];
+
+      // add model flag if we have an override
+      if (modelOverride) {
+        baseArgs.push("--model", modelOverride);
+      }
+
       const cursorArgs = payload.sandbox
-        ? [
-            "--print",
-            fullPrompt,
-            "--output-format",
-            "stream-json",
-            "--approve-mcps",
-            // --force removed in sandbox mode to enforce safety checks
-          ]
-        : ["--print", fullPrompt, "--output-format", "stream-json", "--approve-mcps", "--force"];
+        ? baseArgs // --force removed in sandbox mode to enforce safety checks
+        : [...baseArgs, "--force"];
 
       if (payload.sandbox) {
         log.info("🔒 sandbox mode enabled: restricting to read-only operations");

agents/gemini.ts

@@ -1,6 +1,7 @@
 import { mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";
 import { join } from "node:path";
+import type { Effort } from "../external.ts";
 import { log } from "../utils/cli.ts";
 import { spawn } from "../utils/subprocess.ts";
 import { addInstructions } from "./instructions.ts";
@@ -11,6 +12,15 @@ import {
   installFromGithub,
 } from "./shared.ts";
 
+// effort configuration: model + thinking level
+// thinkingLevel is set via settings.json modelConfig.generateContentConfig.thinkingConfig
+// see: https://ai.google.dev/gemini-api/docs/thinking#thinking-levels
+const geminiEffortConfig: Record<Effort, { model: string; thinkingLevel: string }> = {
+  nothink: { model: "gemini-2.5-flash", thinkingLevel: "LOW" },
+  think: { model: "gemini-2.5-flash", thinkingLevel: "HIGH" },
+  max: { model: "gemini-2.5-pro", thinkingLevel: "HIGH" },
+} as const;
+
 // gemini cli event types inferred from stream-json output (NDJSON format)
 interface GeminiInitEvent {
   type: "init";
@@ -156,8 +166,12 @@ export const gemini = agent({
       ...(githubInstallationToken && { githubInstallationToken }),
     });
   },
-  run: async ({ payload, apiKey, mcpServers, cliPath, repo }) => {
-    configureGeminiMcpServers({ mcpServers, isPublicRepo: repo.isPublic });
+  run: async ({ payload, apiKey, mcpServers, cliPath, repo, effort }) => {
+    // get model and thinking level based on effort
+    const { model, thinkingLevel } = geminiEffortConfig[effort];
+    log.info(`Using model: ${model}, thinkingLevel: ${thinkingLevel}`);
+
+    configureGeminiSettings({ mcpServers, isPublicRepo: repo.isPublic, thinkingLevel });
 
     if (!apiKey) {
       throw new Error("google_api_key or gemini_api_key is required for gemini agent");
@@ -172,6 +186,8 @@ export const gemini = agent({
     if (payload.sandbox) {
       // sandbox mode: read-only tools only
       args = [
+        "--model",
+        model,
         "--allowed-tools",
         "read_file,list_directory,search_file_content,glob,save_memory,write_todos",
         "--allowed-mcp-server-names",
@@ -183,7 +199,7 @@ export const gemini = agent({
     } else {
       // normal mode: --yolo for auto-approval
       // for public repos, shell is excluded via settings.json excludeTools
-      args = ["--yolo", "--output-format=stream-json", "-p", sessionPrompt];
+      args = ["--model", model, "--yolo", "--output-format=stream-json", "-p", sessionPrompt];
       if (repo.isPublic) {
         log.info("🔒 public repo: native shell disabled via excludeTools, using MCP bash");
       }
@@ -276,16 +292,22 @@ export const gemini = agent({
 type ConfigureGeminiParams = {
   mcpServers: ConfigureMcpServersParams["mcpServers"];
   isPublicRepo: boolean;
+  thinkingLevel: string;
 };
 
 /**
- * Configure MCP servers for Gemini by writing to settings.json.
- * Gemini CLI uses `httpUrl` for HTTP/streamable transport, `url` for SSE transport.
- * See: https://github.com/google-gemini/gemini-cli/blob/main/docs/get-started/configuration.md
+ * Configure Gemini CLI settings by writing to settings.json.
+ * - MCP servers: uses `httpUrl` for HTTP/streamable transport
+ * - thinkingLevel: configured via modelConfig.generateContentConfig.thinkingConfig
+ * - For public repos, excludeTools disables native shell
  *
- * For public repos, also configures excludeTools to disable native shell.
+ * See: https://github.com/google-gemini/gemini-cli/blob/main/docs/get-started/configuration.md
  */
-function configureGeminiMcpServers({ mcpServers, isPublicRepo }: ConfigureGeminiParams): void {
+function configureGeminiSettings({
+  mcpServers,
+  isPublicRepo,
+  thinkingLevel,
+}: ConfigureGeminiParams): void {
   const realHome = homedir();
   const geminiConfigDir = join(realHome, ".gemini");
   const settingsPath = join(geminiConfigDir, "settings.json");
@@ -329,17 +351,26 @@ function configureGeminiMcpServers({ mcpServers, isPublicRepo }: ConfigureGemini
     log.info(`Adding MCP server '${serverName}' at ${serverConfig.url}...`);
   }
 
-  // merge with existing settings, overwriting mcpServers
-  // for public repos, exclude native shell tool to prevent secret leakage via env
+  // merge with existing settings, overwriting mcpServers and modelConfig
   const newSettings: Record<string, unknown> = {
     ...existingSettings,
     mcpServers: geminiMcpServers,
+    // configure thinking level via modelConfig
+    // see: https://ai.google.dev/api/generate-content (ThinkingConfig)
+    modelConfig: {
+      generateContentConfig: {
+        thinkingConfig: {
+          thinkingLevel,
+        },
+      },
+    },
   };
 
+  // for public repos, exclude native shell tool to prevent secret leakage via env
   if (isPublicRepo) {
     newSettings.excludeTools = ["run_shell_command"];
   }
 
   writeFileSync(settingsPath, JSON.stringify(newSettings, null, 2), "utf-8");
-  log.info(`» MCP config written to ${settingsPath}`);
+  log.info(`» Gemini settings written to ${settingsPath}`);
 }

agents/opencode.ts

@@ -22,7 +22,15 @@ export const opencode = agent({
       installDependencies: true,
     });
   },
-  run: async ({ payload, apiKey: _apiKey, apiKeys, mcpServers, cliPath, repo }) => {
+  run: async ({
+    payload,
+    apiKey: _apiKey,
+    apiKeys,
+    mcpServers,
+    cliPath,
+    repo,
+    effort: _effort,
+  }) => {
     // 1. configure home/config directory
     const tempHome = process.env.PULLFROG_TEMP_DIR!;
     const configDir = join(tempHome, ".config", "opencode");
@@ -60,10 +68,9 @@ export const opencode = agent({
 
     // add API keys from apiKeys object
     for (const [key, value] of Object.entries(apiKeys || {})) {
-      const upperKey = key.toUpperCase();
-      env[upperKey] = value;
+      env[key.toUpperCase()] = value;
       // also set GOOGLE_GENERATIVE_AI_API_KEY for Google provider compatibility
-      if (upperKey === "GEMINI_API_KEY") {
+      if (key === "GEMINI_API_KEY") {
         env.GOOGLE_GENERATIVE_AI_API_KEY = value;
       }
     }

agents/shared.ts

@@ -6,7 +6,13 @@ import { join } from "node:path";
 import { pipeline } from "node:stream/promises";
 import type { McpHttpServerConfig } from "@anthropic-ai/claude-agent-sdk";
 import type { show } from "@ark/util";
-import { type AgentManifest, type AgentName, agentsManifest, type Payload } from "../external.ts";
+import {
+  type AgentManifest,
+  type AgentName,
+  agentsManifest,
+  type Effort,
+  type Payload,
+} from "../external.ts";
 import { log } from "../utils/cli.ts";
 import { getGitHubInstallationToken } from "../utils/github.ts";
 
@@ -40,6 +46,7 @@ export interface AgentConfig {
   mcpServers: Record<string, McpHttpServerConfig>;
   cliPath: string;
   repo: RepoInfo;
+  effort: Effort;
 }
 
 /**

docs/webfetch.md

@@ -0,0 +1,520 @@
+# WebFetch Tool Analysis
+
+Analysis of webfetch/URL fetching implementations across three AI coding agents to inform the design of pullfrog's custom webfetch MCP tool.
+
+---
+
+## 1. OpenCode Implementation
+
+**Source**: `packages/opencode/src/tool/webfetch.ts`
+
+### Architecture
+
+OpenCode's webfetch is straightforward - a simple fetch wrapper with HTML-to-markdown conversion:
+
+```typescript
+const response = await fetch(params.url, {
+  signal: AbortSignal.any([controller.signal, ctx.abort]),
+  headers: {
+    "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36...",
+    Accept: acceptHeader,
+    "Accept-Language": "en-US,en;q=0.9",
+  },
+})
+```
+
+### Key Features
+
+| Feature | Implementation |
+|---------|---------------|
+| **Output formats** | `text`, `markdown`, `html` (default: markdown) |
+| **HTML→Markdown** | Uses `turndown` library |
+| **Max response size** | 5MB hard limit |
+| **Timeout** | 30s default, 120s max |
+| **Permission system** | Application-level `ctx.ask()` prompt |
+| **Domain blocking** | None - relies on user approval |
+| **Caching** | None |
+| **Redirect handling** | Native fetch behavior |
+
+### HTML Processing
+
+Two methods depending on output format:
+
+1. **`extractTextFromHTML()`** - Uses Bun's `HTMLRewriter` to strip scripts/styles and extract text
+2. **`convertHTMLToMarkdown()`** - Uses `turndown` with sensible defaults (ATX headings, fenced code blocks)
+
+### Permission Model
+
+```typescript
+await ctx.ask({
+  permission: "webfetch",
+  patterns: [params.url],
+  always: ["*"],  // User can allow all future requests
+  metadata: { url, format, timeout },
+})
+```
+
+**Verdict**: No enforcement - purely advisory. If user approves, the fetch proceeds with no restrictions.
+
+### What I Like
+- Clean, minimal implementation
+- Good HTML processing with `turndown`
+- Sensible size limits (5MB)
+- Format flexibility
+
+### What I Don't Like
+- No domain whitelisting/blocklisting
+- No caching (repeated requests to same URL are wasteful)
+- Permission system is advisory-only
+- No redirect safety checks
+
+---
+
+## 2. Claude Code Implementation
+
+**Source**: Extracted from bundled `claude` CLI binary
+
+### Architecture
+
+Claude Code uses a more sophisticated approach with server-side domain validation:
+
+```javascript
+// Domain validation before fetch
+async function Ci5(domain) {
+  const response = await fetch(
+    `https://claude.ai/api/web/domain_info?domain=${encodeURIComponent(domain)}`
+  );
+  if (response.status === 200) {
+    return response.data.can_fetch === true 
+      ? { status: "allowed" } 
+      : { status: "blocked" };
+  }
+  return { status: "check_failed" };
+}
+```
+
+### Key Features
+
+| Feature | Implementation |
+|---------|---------------|
+| **Domain blocklist** | Server-side API at `claude.ai/api/web/domain_info` |
+| **Permission format** | `WebFetch(domain:example.com)` - domain-only, not URLs |
+| **Wildcard support** | `domain:*.google.com` patterns |
+| **HTTP→HTTPS upgrade** | Automatic protocol upgrade |
+| **Caching** | 15-minute self-cleaning cache |
+| **HTML→Markdown** | Uses `turndown` |
+| **Redirect handling** | Special handling - informs user of cross-host redirects |
+| **Enterprise override** | `skipWebFetchPreflight` setting |
+
+### Domain Permission Model
+
+Claude Code enforces domain-level permissions, not URL-level:
+
+```javascript
+WebFetch: (A) => {
+  if (A.includes("://") || A.startsWith("http"))
+    return {
+      valid: false,
+      error: "WebFetch permissions use domain format, not URLs",
+      suggestion: 'Use "domain:hostname" format',
+      examples: ["WebFetch(domain:example.com)", "WebFetch(domain:github.com)"]
+    };
+  if (!A.startsWith("domain:"))
+    return {
+      valid: false,
+      error: 'WebFetch permissions must use "domain:" prefix',
+      examples: ["WebFetch(domain:example.com)", "WebFetch(domain:*.google.com)"]
+    };
+  return { valid: true };
+}
+```
+
+### Blocklist Enforcement Flow
+
+```
+User requests URL
+       ↓
+Extract hostname
+       ↓
+Check claude.ai/api/web/domain_info?domain=hostname
+       ↓
+┌──────────────────────────────────┐
+│ allowed → proceed with fetch     │
+│ blocked → throw AC0 error        │
+│ check_failed → throw QC0 error   │
+└──────────────────────────────────┘
+```
+
+### Redirect Handling
+
+When a URL redirects to a different host:
+```javascript
+// Returns special response asking user to manually re-request
+return `To complete your request, I need to fetch content from the redirected URL. 
+Please use WebFetch again with these parameters:
+- url: "${redirectUrl}"
+- prompt: "${originalPrompt}"`;
+```
+
+This prevents open redirect attacks where a "safe" domain redirects to a malicious one.
+
+### What I Like
+- **Server-side blocklist** - centralized, updateable without client changes
+- **Domain-level permissions** - prevents path-based bypasses
+- **Redirect safety** - cross-host redirects require explicit user action
+- **15-minute caching** - reduces redundant requests
+- **Enterprise override** - `skipWebFetchPreflight` for corporate environments
+
+### What I Don't Like
+- **External dependency** - requires `claude.ai` API to be available
+- **No local blocklist** - can't work offline or with custom blocklists
+- **Opaque blocklist** - users can't see what's blocked or why
+
+---
+
+## 3. Gemini CLI Implementation
+
+**Source**: `@google/gemini-cli` npm package
+
+### Architecture
+
+Gemini CLI takes a fundamentally different approach - it doesn't have a dedicated webfetch tool. Instead it relies on:
+
+1. **Google Search grounding** - built into the Gemini API
+2. **MCP servers** - external tools can provide fetch capabilities
+3. **No native URL fetching** - by design
+
+### Key Observations
+
+From searching the codebase:
+- No `webfetch`, `url_fetch`, or similar tool definitions
+- Has `github_fetch.ts` for fetching GitHub releases (internal use)
+- Relies on model's built-in capabilities or MCP extensions
+
+### Why No WebFetch?
+
+Gemini's design philosophy appears to be:
+1. Use the model's grounding capabilities for web information
+2. Delegate specialized fetching to MCP servers
+3. Avoid building network access into the CLI itself
+
+### What I Like
+- **Clean separation** - network access is opt-in via MCP
+- **Security by default** - no built-in way to exfiltrate data
+
+### What I Don't Like
+- **Missing functionality** - can't fetch arbitrary URLs
+- **Requires MCP setup** - more complex for users who need fetching
+
+---
+
+## 4. Pullfrog Design Decisions
+
+### Core Requirements
+
+1. **Domain-level whitelisting** - enforced in-tool, not advisory
+2. **Simple implementation** - no external API dependencies
+3. **GitHub-focused** - optimized for common development URLs
+
+### Proposed Architecture
+
+```
+┌─────────────────────────────────────────────────────┐
+│                   WebFetch Tool                      │
+├─────────────────────────────────────────────────────┤
+│  1. Parse URL → extract hostname                     │
+│  2. Check against DOMAIN_ALLOWLIST                   │
+│  3. If not allowed → return error (not throw)        │
+│  4. Fetch with timeout + size limits                 │
+│  5. Convert HTML → Markdown if needed                │
+│  6. Return content                                   │
+└─────────────────────────────────────────────────────┘
+```
+
+### Domain Allowlist Strategy
+
+**Included in initial allowlist**:
+```typescript
+const DOMAIN_ALLOWLIST = new Set([
+  // Documentation sites
+  "docs.github.com",
+  "developer.mozilla.org",
+  "nodejs.org",
+  "docs.python.org",
+  "go.dev",
+  "doc.rust-lang.org",
+  "docs.microsoft.com",
+  "learn.microsoft.com",
+  
+  // Package registries (documentation)
+  "npmjs.com",
+  "www.npmjs.com",
+  "pypi.org",
+  "crates.io",
+  "pkg.go.dev",
+  
+  // GitHub (raw content, gists)
+  "raw.githubusercontent.com",
+  "gist.githubusercontent.com",
+  
+  // Common API documentation
+  "api.github.com",  // Already have GitHub tools, but for reference docs
+]);
+```
+
+**Explicitly NOT included**:
+- `github.com` itself - we have dedicated GitHub MCP tools
+- Social media sites
+- General web pages
+- Arbitrary user-provided domains
+
+### Features Borrowed from Each Agent
+
+| Feature | Source | Included? | Rationale |
+|---------|--------|-----------|-----------|
+| HTML→Markdown via turndown | OpenCode | ✅ | Clean, proven library |
+| 5MB size limit | OpenCode | ✅ | Sensible default |
+| Domain-level permissions | Claude Code | ✅ | Core requirement |
+| Redirect safety checks | Claude Code | ✅ | Prevents open redirect attacks |
+| 15-minute caching | Claude Code | ❌ | Adds complexity, MCP is stateless |
+| Server-side blocklist | Claude Code | ❌ | External dependency |
+| Enterprise override | Claude Code | ❌ | Not needed for GitHub Actions |
+| No built-in fetching | Gemini | ❌ | We need this functionality |
+
+### Features NOT Included (and why)
+
+1. **Caching** - MCP tools are stateless by design. Caching would require shared state across requests. The agent can cache results itself.
+
+2. **Server-side blocklist** - Would require standing up an API endpoint. The allowlist approach is simpler and more transparent.
+
+3. **User permission prompts** - In GitHub Actions context, there's no interactive user. Allowlist is enforced automatically.
+
+4. **Wildcard domain patterns** - Adds complexity. Start with explicit domains, add patterns if needed.
+
+5. **Multiple output formats** - Start with markdown only. Can add `text` and `html` later if needed.
+
+### Error Handling Strategy
+
+Unlike OpenCode/Claude which throw errors, we return errors as content:
+
+```typescript
+// Domain not allowed - return message, don't throw
+if (!isDomainAllowed(hostname)) {
+  return {
+    output: `Domain "${hostname}" is not in the allowlist. Allowed domains: ${Array.from(DOMAIN_ALLOWLIST).join(", ")}`,
+    error: true,
+  };
+}
+```
+
+This lets the agent understand the limitation and potentially find alternative approaches.
+
+### Redirect Handling
+
+Adopt Claude Code's approach with modification:
+
+```typescript
+// If redirect crosses domains, check the new domain
+if (response.redirected) {
+  const redirectUrl = new URL(response.url);
+  if (!isDomainAllowed(redirectUrl.hostname)) {
+    return {
+      output: `URL redirected to "${redirectUrl.hostname}" which is not in the allowlist.`,
+      error: true,
+    };
+  }
+}
+```
+
+---
+
+## 5. Implementation Plan
+
+### Summary
+
+Add a new `webfetch` MCP tool that fetches web content with domain-level whitelisting enforced server-side. The whitelist is configured via the payload (from GitHub App), and non-whitelisted domains return a helpful message guiding the LLM to alternative approaches.
+
+### Key Design Decisions
+
+| Aspect | OpenCode | Claude Code | Our Implementation |
+|--------|----------|-------------|-------------------|
+| **Whitelisting** | Permission prompt (advisory) | External API `domain_info` | Payload-configured whitelist |
+| **Enforcement** | None (user approval) | Server-side check | Server-side check |
+| **HTML Processing** | Turndown for markdown | Turndown for markdown | Turndown for markdown |
+| **Redirects** | Follows automatically | Detects cross-host redirects | Follow with host check |
+| **Timeout** | 30s default, 120s max | Configurable | 30s default, 120s max |
+
+### Step 1: Add whitelist to payload type
+
+Update `index.ts` to include `allowedWebFetchDomains`:
+
+```typescript
+interface Payload {
+  // ... existing fields
+  allowedWebFetchDomains?: string[];  // e.g. ["github.com", "*.npmjs.com", "docs.python.org"]
+}
+```
+
+### Step 2: Create `mcp/webfetch.ts`
+
+```typescript
+// Core structure
+export const WebFetchParams = type({
+  url: "string",
+  "format?": "'markdown' | 'text' | 'html'",
+  "timeout?": "number",
+});
+
+export function WebFetchTool(ctx: ToolContext) {
+  return tool({
+    name: "webfetch",
+    description: `Fetch content from whitelisted web URLs...`,
+    parameters: WebFetchParams,
+    execute: execute(async (params) => {
+      // 1. Validate URL format
+      // 2. Check domain against whitelist (from ctx.payload)
+      // 3. Fetch with timeout and size limits
+      // 4. Convert HTML to markdown if needed
+      // 5. Return content or guidance message
+    }),
+  });
+}
+```
+
+### Step 3: Domain Matching Logic
+
+Support wildcards for subdomains:
+
+- `github.com` - exact match
+- `*.github.com` - any subdomain (e.g., `docs.github.com`, `api.github.com`)
+- `*.npmjs.com` - matches `www.npmjs.com`, `registry.npmjs.com`, etc.
+
+```typescript
+function isDomainAllowed(hostname: string, whitelist: string[]): boolean {
+  for (const pattern of whitelist) {
+    if (pattern.startsWith("*.")) {
+      const suffix = pattern.slice(1); // ".github.com"
+      if (hostname.endsWith(suffix) || hostname === pattern.slice(2)) {
+        return true;
+      }
+    } else if (hostname === pattern) {
+      return true;
+    }
+  }
+  return false;
+}
+```
+
+### Step 4: Response for Non-Whitelisted Domains
+
+When domain is not whitelisted, return guidance (not an error):
+
+```typescript
+return {
+  allowed: false,
+  message: `The domain "${hostname}" is not in the allowed list for direct fetching. ` +
+    `Consider using web_search to find relevant information, or ask the user to ` +
+    `provide the content directly. Allowed domains: ${whitelist.join(", ")}`,
+};
+```
+
+### Step 5: HTML to Markdown Conversion
+
+Use Turndown (same as OpenCode) for HTML-to-markdown conversion:
+
+```typescript
+import TurndownService from "turndown";
+
+function htmlToMarkdown(html: string): string {
+  const turndown = new TurndownService({
+    headingStyle: "atx",
+    codeBlockStyle: "fenced",
+  });
+  turndown.remove(["script", "style", "meta", "link"]);
+  return turndown.turndown(html);
+}
+```
+
+### Step 6: Register the Tool
+
+Add to `mcp/index.ts`:
+
+```typescript
+import { WebFetchTool } from "./webfetch.ts";
+
+// In the tools array
+WebFetchTool(ctx),
+```
+
+### Data Flow
+
+```mermaid
+sequenceDiagram
+    participant LLM
+    participant MCP as MCP Server
+    participant WF as WebFetch Tool
+    participant Web as External URL
+
+    LLM->>MCP: webfetch(url, format)
+    MCP->>WF: execute(params)
+    WF->>WF: Parse URL, extract hostname
+    WF->>WF: Check whitelist from payload
+    alt Domain allowed
+        WF->>Web: fetch(url)
+        Web-->>WF: Response
+        WF->>WF: Convert to markdown
+        WF-->>MCP: {content, contentType}
+        MCP-->>LLM: Success result
+    else Domain not allowed
+        WF-->>MCP: {allowed: false, guidance}
+        MCP-->>LLM: Guidance message
+    end
+```
+
+### Files to Create/Modify
+
+| File | Action |
+|------|--------|
+| `mcp/webfetch.ts` | Create - main tool implementation |
+| `mcp/index.ts` | Modify - register the tool |
+| `index.ts` | Modify - add `allowedWebFetchDomains` to payload type |
+| `package.json` | Modify - add `turndown` dependency |
+
+### Dependencies
+
+Add to `package.json`:
+
+- `turndown` - HTML to markdown conversion (same as OpenCode)
+- `@types/turndown` - TypeScript types
+
+---
+
+## 6. Implementation Checklist
+
+- [ ] Add `allowedWebFetchDomains` field to payload type in `index.ts`
+- [ ] Create `mcp/webfetch.ts` with domain whitelisting and HTML conversion
+- [ ] Register `WebFetchTool` in `mcp/index.ts`
+- [ ] Add `turndown` and `@types/turndown` dependencies to `package.json`
+- [ ] Test with allowed domains
+- [ ] Test with blocked domains
+- [ ] Test redirect behavior
+
+---
+
+## 7. Open Questions
+
+1. **Should we support query parameters in allowlist?** 
+   - e.g., allow `api.example.com/v1/*` but not `api.example.com/admin/*`
+   - Initial decision: No, domain-level only
+
+2. **Should we allow configurable allowlists?**
+   - Via environment variable or config file?
+   - Initial decision: No, hardcoded for simplicity
+
+3. **Should we support authentication headers?**
+   - For private documentation sites
+   - Initial decision: No, security risk
+
+4. **Rate limiting?**
+   - Prevent agent from hammering a site
+   - Initial decision: Rely on timeout, add if needed

docs/websearch.md

@@ -0,0 +1,435 @@
+# Web Search Functionality by Agent
+
+This document describes how each supported agent implements web search functionality.
+
+## Summary
+
+| Agent | Tool Name | Search Provider | API/Method |
+|-------|-----------|-----------------|------------|
+| Claude Code | `WebSearch` | Anthropic internal | Claude Code SDK |
+| Gemini CLI | `google_web_search` | Google Search via Gemini API | `generateContent` with `model: 'web-search'` |
+| OpenCode | `websearch` | Exa AI | MCP protocol to `https://mcp.exa.ai/mcp` |
+
+All three agents also support a separate **web fetch** tool for directly retrieving and parsing web page content.
+
+---
+
+## Claude Code
+
+### Tools
+- `WebSearch` - Search the web for information
+- `WebFetch` - Fetch and process web content
+
+### Implementation
+Native functionality through `@anthropic-ai/claude-agent-sdk` (closed source). The actual search provider is internal to Anthropic's infrastructure.
+
+### Configuration in Pullfrog
+
+Web search can be disabled via the `disallowedTools` option:
+
+```typescript
+// In sandbox mode, web tools are disabled
+disallowedTools: ["Bash", "WebSearch", "WebFetch", "Write"]
+```
+
+---
+
+## Gemini CLI
+
+### Tools
+- `google_web_search` - Perform web searches using Google Search
+- `web_fetch` - Fetch and process content from URLs
+
+### Implementation
+
+Source: [`packages/core/src/tools/web-search.ts`](https://github.com/google-gemini/gemini-cli/blob/main/packages/core/src/tools/web-search.ts)
+
+**How it works:**
+1. Sends query to Gemini API using `generateContent` with `model: 'web-search'`
+2. Google performs the search and returns results with grounding metadata
+3. Response includes inline citations, source URLs, and titles
+
+```typescript
+const response = await geminiClient.generateContent(
+  { model: 'web-search' },
+  [{ role: 'user', parts: [{ text: this.params.query }] }],
+  signal,
+);
+```
+
+### Features
+- Returns processed summary (not raw search results)
+- Inline citations with grounding metadata
+- Sources list with titles and URIs
+- UTF-8 byte position handling for accurate citation insertion
+
+### Parameters
+- `query` (string, required): The search query
+
+### Web Fetch
+
+The `web_fetch` tool processes content from URLs:
+- Uses Gemini API's `urlContext` feature
+- Fallback to direct HTTP fetch with `html-to-text` conversion
+- Supports up to 20 URLs per request
+- Converts GitHub blob URLs to raw URLs automatically
+
+---
+
+## OpenCode
+
+### Tools
+- `websearch` - Search the web using Exa AI
+- `webfetch` - Fetch and read web pages
+
+### Implementation
+
+Source: [`packages/opencode/src/tool/websearch.ts`](https://github.com/sst/opencode/blob/main/packages/opencode/src/tool/websearch.ts)
+
+**How it works:**
+1. Calls Exa AI's MCP endpoint at `https://mcp.exa.ai/mcp`
+2. Uses JSON-RPC protocol to invoke the `web_search_exa` tool
+3. Parses SSE response for search results
+
+```typescript
+const searchRequest: McpSearchRequest = {
+  jsonrpc: "2.0",
+  id: 1,
+  method: "tools/call",
+  params: {
+    name: "web_search_exa",
+    arguments: {
+      query: params.query,
+      type: params.type || "auto",
+      numResults: params.numResults || 8,
+      livecrawl: params.livecrawl || "fallback",
+      contextMaxCharacters: params.contextMaxCharacters,
+    },
+  },
+}
+```
+
+### Features
+- Real-time web searches with content scraping
+- Configurable result count (default: 8)
+- Live crawl modes: `fallback` (backup if cached unavailable) or `preferred` (prioritize live crawling)
+- Search types: `auto` (balanced), `fast` (quick results), `deep` (comprehensive)
+- Context max characters for LLM optimization
+
+### Parameters
+- `query` (string, required): The search query
+- `numResults` (number, optional): Number of results to return (default: 8)
+- `livecrawl` (enum, optional): `"fallback"` | `"preferred"`
+- `type` (enum, optional): `"auto"` | `"fast"` | `"deep"`
+- `contextMaxCharacters` (number, optional): Maximum characters for context
+
+### Configuration in Pullfrog
+
+Web tools are configured via the permission config in `opencode.json`:
+
+```typescript
+// In sandbox mode
+permission: {
+  webfetch: "deny",
+  // ...
+}
+
+// In normal mode
+permission: {
+  webfetch: "allow",
+  // ...
+}
+```
+
+### Environment Variables
+- `OPENCODE_ENABLE_EXA` - Enable Exa web search tools (required for "zen" users)
+
+### Web Fetch
+
+The `webfetch` tool directly fetches URLs:
+- Direct HTTP fetch with browser-like User-Agent
+- HTML to Markdown conversion using Turndown
+- Configurable timeout (max 120 seconds)
+- 5MB response size limit
+
+---
+
+## Comparison
+
+| Feature | Claude Code | Gemini CLI | OpenCode |
+|---------|-------------|------------|----------|
+| Search Provider | Anthropic | Google | Exa AI |
+| Result Format | Summary | Summary + Citations | Raw content |
+| URL Fetching | Yes (`WebFetch`) | Yes (`web_fetch`) | Yes (`webfetch`) |
+| Grounding/Citations | Unknown | Yes | No |
+| Configurable Results | No | No | Yes (numResults) |
+| Search Depth Options | No | No | Yes (auto/fast/deep) |
+| Live Crawling | Unknown | Fallback only | Configurable |
+
+---
+
+## Security Considerations
+
+In Pullfrog's sandbox mode:
+- **Claude Code**: `WebSearch` and `WebFetch` are explicitly disabled via `disallowedTools`
+- **Gemini CLI**: No explicit disable mechanism in the wrapper (relies on default behavior)
+- **OpenCode**: `webfetch` permission set to `"deny"` in sandbox mode
+
+For public repositories, consider the implications of web search/fetch:
+- Fetched content could potentially be used to inject prompts
+- Search queries might leak information about the codebase context
+
+---
+
+## Proposed Implementation Plan
+
+### Option 1: Use Native Agent Web Search (Current State)
+
+Each agent uses its own built-in web search:
+- **Pros**: No additional implementation, leverages each provider's strengths
+- **Cons**: Inconsistent behavior across agents, no unified control
+
+**Current gaps:**
+- Gemini CLI has no explicit disable mechanism for web search in sandbox mode
+- No unified way to configure web search across all agents
+
+### Option 2: Unified MCP Web Search Tool
+
+Add a `web_search` tool to the Pullfrog MCP server (`mcp/`) that all agents can use:
+
+```
+mcp/
+├── bash.ts
+├── webSearch.ts    # New unified web search tool
+└── ...
+```
+
+**Implementation approach:**
+
+1. **Create `mcp/webSearch.ts`** with a provider-agnostic interface:
+   ```typescript
+   export const webSearchTool = {
+     name: "web_search",
+     description: "Search the web for information",
+     inputSchema: {
+       type: "object",
+       properties: {
+         query: { type: "string", description: "Search query" },
+         numResults: { type: "number", description: "Number of results (default: 5)" },
+       },
+       required: ["query"],
+     },
+   };
+   ```
+
+2. **Choose a search provider** (options):
+   - **Exa AI** - Already used by OpenCode, good LLM-optimized results
+   - **Tavily** - Popular for AI agents, provides search + content extraction
+   - **SerpAPI** - Google results via API
+   - **Brave Search API** - Privacy-focused alternative
+
+3. **Add to MCP server** in `mcp/server.ts`:
+   ```typescript
+   import { webSearchTool, handleWebSearch } from "./webSearch.ts";
+   // Register tool...
+   ```
+
+4. **Disable native web search** for each agent:
+   - Claude: Add `"WebSearch"` to `disallowedTools`
+   - Gemini: Add `"google_web_search"` to `excludeTools` in settings.json
+   - OpenCode: Set `websearch: "deny"` in permission config
+
+**Pros:**
+- Consistent behavior across all agents
+- Centralized control for security/sandbox modes
+- Can filter/sanitize results before returning to agent
+- Single API key management
+
+**Cons:**
+- Additional API costs (search provider)
+- Loses provider-specific features (e.g., Gemini's grounding metadata)
+
+### Option 3: Hybrid Approach
+
+Allow native web search for private repos, use MCP tool for public repos:
+
+```typescript
+// In agent configuration
+const useNativeWebSearch = !repo.isPublic;
+
+// Claude
+disallowedTools: repo.isPublic ? ["WebSearch", "WebFetch"] : [];
+
+// Gemini  
+excludeTools: repo.isPublic ? ["google_web_search", "web_fetch"] : [];
+
+// OpenCode
+permission: {
+  websearch: repo.isPublic ? "deny" : "allow",
+}
+```
+
+Then for public repos, agents would use the MCP `web_search` tool which:
+- Filters sensitive queries
+- Sanitizes returned content
+- Logs all searches for audit
+
+### Recommended Approach
+
+**Short-term**: Implement Option 3 (Hybrid) with these steps:
+
+1. [ ] Add `excludeTools: ["google_web_search"]` for Gemini in public repo mode
+2. [ ] Ensure OpenCode `websearch` permission is properly set for sandbox mode
+3. [ ] Document the current native web search behavior for each agent
+
+**Medium-term**: Implement Option 2 (Unified MCP) for public repos:
+
+1. [ ] Create `mcp/webSearch.ts` using Exa AI (consistent with OpenCode)
+2. [ ] Add `EXA_API_KEY` to secrets handling
+3. [ ] Register web search in MCP server
+4. [ ] Disable native web search for all agents when MCP tool is available
+5. [ ] Add result sanitization to prevent prompt injection
+
+### API Key Requirements
+
+| Provider | Environment Variable | Notes |
+|----------|---------------------|-------|
+| Exa AI | `EXA_API_KEY` | Already used by OpenCode |
+| Tavily | `TAVILY_API_KEY` | Popular alternative |
+| Brave | `BRAVE_API_KEY` | Privacy-focused |
+
+For the unified MCP approach, only one search provider API key would be needed.
+
+---
+
+## Proposed Implementation Plan
+
+### Option A: Unified MCP Web Search Tool
+
+Create a custom MCP tool that provides consistent web search across all agents.
+
+**Pros:**
+- Consistent behavior and results across agents
+- Full control over search provider and rate limiting
+- Can implement caching and deduplication
+- Single point for security filtering
+
+**Cons:**
+- Additional infrastructure (need a search API key)
+- Latency from proxying through MCP server
+
+**Implementation:**
+1. Add `websearch` tool to `mcp/` directory
+2. Integrate with a search provider (options: Exa AI, SerpAPI, Brave Search, Tavily)
+3. Configure each agent to use MCP tool instead of native:
+   - Claude: Add to `disallowedTools` and provide via MCP
+   - Gemini: Use `excludeTools` in settings.json for `google_web_search`
+   - OpenCode: Disable native via permission config
+
+```typescript
+// mcp/websearch.ts
+export const websearchTool = {
+  name: "websearch",
+  description: "Search the web for current information",
+  inputSchema: {
+    type: "object",
+    properties: {
+      query: { type: "string", description: "Search query" },
+      numResults: { type: "number", description: "Number of results (1-10)" },
+    },
+    required: ["query"],
+  },
+  handler: async ({ query, numResults = 5 }) => {
+    // Use Exa, Brave, or other search API
+    const results = await searchProvider.search(query, numResults);
+    return formatResults(results);
+  },
+};
+```
+
+### Option B: Native Tools with Configuration
+
+Keep using each agent's native web search but add consistent configuration.
+
+**Pros:**
+- No additional infrastructure
+- Agents can use optimized native implementations
+- Less latency
+
+**Cons:**
+- Inconsistent results across agents
+- Different capabilities per agent
+- Harder to control/audit searches
+
+**Implementation:**
+1. Add `websearch_enabled` option to payload/config
+2. Update each agent wrapper:
+   - Claude: Toggle `WebSearch` in `disallowedTools`
+   - Gemini: Add `google_web_search` to `excludeTools` in settings.json
+   - OpenCode: Set `websearch` permission in config
+
+```typescript
+// agents/claude.ts
+const disallowedTools = payload.websearchEnabled 
+  ? ["Bash"] 
+  : ["Bash", "WebSearch", "WebFetch"];
+
+// agents/gemini.ts  
+if (!payload.websearchEnabled) {
+  newSettings.excludeTools = [...(newSettings.excludeTools || []), "google_web_search"];
+}
+
+// agents/opencode.ts
+permission: {
+  websearch: payload.websearchEnabled ? "allow" : "deny",
+  // ...
+}
+```
+
+### Option C: Hybrid Approach (Recommended)
+
+Use native tools when available, with MCP fallback for consistency.
+
+**Implementation:**
+1. Define a `websearch` MCP tool as fallback
+2. For agents with good native search (Claude, Gemini): use native
+3. For agents without (or with unreliable) search: use MCP tool
+4. Add configuration to force MCP-only mode if needed
+
+```typescript
+// Per-agent configuration
+const agentWebSearchConfig = {
+  claude: { useNative: true, mcpFallback: false },
+  gemini: { useNative: true, mcpFallback: false },
+  opencode: { useNative: false, mcpFallback: true }, // Exa requires API key
+};
+```
+
+### Required Changes by Option
+
+| Change | Option A | Option B | Option C |
+|--------|----------|----------|----------|
+| New MCP tool | Yes | No | Yes |
+| Search API key | Yes | No | Optional |
+| Agent wrapper changes | Yes | Yes | Yes |
+| Action input changes | No | Yes | Yes |
+| External dependencies | Yes | No | Optional |
+
+### Recommended Next Steps
+
+1. **Decide on search provider** - If going with MCP approach:
+   - Exa AI: Already used by OpenCode, good for code-related searches
+   - Brave Search: Privacy-focused, good general search
+   - Tavily: Designed for AI agents, includes content extraction
+
+2. **Add configuration** - New action inputs:
+   ```yaml
+   websearch:
+     description: 'Enable web search functionality'
+     required: false
+     default: 'false'
+   ```
+
+3. **Implement per-agent** - Start with Option B (simplest), upgrade to C if needed
+
+4. **Add security controls** - Query filtering, domain allowlists, rate limiting

effort.md

@@ -0,0 +1,92 @@
+# Effort Levels
+
+Pullfrog supports three effort levels that control model selection and reasoning depth:
+
+- **`nothink`** — Fast, minimal reasoning. Best for simple tasks.
+- **`think`** — Balanced (default). Good for most tasks.
+- **`max`** — Maximum capability. Best for complex tasks requiring deep reasoning.
+
+The effort level can be specified via the `effort` input in `action.yml` or in the payload's `effort` field.
+
+---
+
+## Claude Code
+
+Claude Code uses model selection based on effort level.
+
+| Effort | Model | Description |
+|--------|-------|-------------|
+| `nothink` | `haiku` | Fast, efficient |
+| `think` | `opusplan` | Opus for planning, Sonnet for execution |
+| `max` | `opus` | Full Opus |
+
+> **Future direction:** Anthropic's beta `effort` parameter (`low`/`medium`/`high`) could replace model selection, using Opus 4.5 for all tasks with effort controlling token spend. See [Anthropic Effort Docs](https://platform.claude.com/docs/en/build-with-claude/effort).
+
+---
+
+## Codex (OpenAI)
+
+Codex uses both model selection and the `modelReasoningEffort` parameter from `ThreadOptions`.
+
+| Effort | Model | `modelReasoningEffort` | Description |
+|--------|-------|------------------------|-------------|
+| `nothink` | `gpt-5.1-codex-mini` | `"low"` | Smaller model, reduced reasoning |
+| `think` | `gpt-5.1-codex` | default | Standard model, default reasoning |
+| `max` | `gpt-5.1-codex-max` | `"high"` | Largest model, maximum reasoning |
+
+Valid values for `modelReasoningEffort`: `"minimal"` | `"low"` | `"medium"` | `"high"`
+
+Reference: [Codex Config Reference](https://developers.openai.com/codex/config-reference/)
+
+---
+
+## Gemini
+
+Gemini uses a combination of model selection and `thinkingLevel` configuration via `settings.json`.
+
+| Effort | Model | `thinkingLevel` | Description |
+|--------|-------|-----------------|-------------|
+| `nothink` | `gemini-2.5-flash` | `LOW` | Fast model, minimal thinking |
+| `think` | `gemini-2.5-flash` | `HIGH` | Fast model, deep thinking |
+| `max` | `gemini-2.5-pro` | `HIGH` | Most capable model, deep thinking |
+
+The `thinkingLevel` is configured via:
+```json
+{
+  "modelConfig": {
+    "generateContentConfig": {
+      "thinkingConfig": {
+        "thinkingLevel": "LOW"
+      }
+    }
+  }
+}
+```
+
+Reference: [Gemini Thinking Docs](https://ai.google.dev/gemini-api/docs/thinking#thinking-levels)
+
+---
+
+## Cursor
+
+Cursor uses model selection via the `--model` CLI flag. Project-level configuration in `.cursor/cli.json` takes precedence if a `model` is specified there.
+
+| Effort | Model | Description |
+|--------|-------|-------------|
+| `nothink` | `auto` (default) | Let Cursor select optimal model |
+| `think` | `auto` (default) | Let Cursor select optimal model |
+| `max` | `opus-4.5-thinking` | Claude 4.5 Opus with thinking |
+
+**Note:** If the project has `.cursor/cli.json` with a `model` field, that model is used regardless of effort level.
+
+---
+
+## OpenCode
+
+OpenCode does not currently have affordances for effort-level configuration. The effort parameter is ignored.
+
+| Effort | Behavior |
+|--------|----------|
+| `nothink` | No effect |
+| `think` | No effect |
+| `max` | No effect |

entry.ts

@@ -5,27 +5,23 @@
  */
 
 import * as core from "@actions/core";
-import { flatMorph } from "@ark/util";
-import { agents } from "./agents/index.ts";
-import { type Inputs, main } from "./main.ts";
+import { Inputs, main } from "./main.ts";
 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)
+  // Change to cwd input or GITHUB_WORKSPACE (where actions/checkout puts the 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);
-    log.debug(`New working directory: ${process.cwd()}`);
+  const cwd = core.getInput("cwd") || process.env.GITHUB_WORKSPACE;
+  if (cwd && process.cwd() !== cwd) {
+    log.debug(`changing to working directory: ${cwd}`);
+    process.chdir(cwd);
   }
 
   try {
-    const inputs: Required<Inputs> = {
+    const inputs = Inputs.assert({
       prompt: core.getInput("prompt", { required: true }),
-      ...flatMorph(agents, (_, agent) =>
-        agent.apiKeyNames.map((inputKey) => [inputKey, core.getInput(inputKey)])
-      ),
-    };
+      effort: core.getInput("effort") || "think",
+    });
 
     const result = await main(inputs);
 

esbuild.config.js

@@ -67,4 +67,12 @@ await build({
   plugins: [stripShebangPlugin],
 });
 
+// Build the get-installation-token action
+await build({
+  ...sharedConfig,
+  entryPoints: ["./get-installation-token/entry.ts"],
+  outfile: "./get-installation-token/entry",
+  plugins: [stripShebangPlugin],
+});
+
 console.log("✅ Build completed successfully!");

external.ts

@@ -20,22 +20,22 @@ export interface AgentManifest {
 export const agentsManifest = {
   claude: {
     displayName: "Claude Code",
-    apiKeyNames: ["anthropic_api_key"],
+    apiKeyNames: ["ANTHROPIC_API_KEY"],
     url: "https://claude.com/claude-code",
   },
   codex: {
     displayName: "Codex CLI",
-    apiKeyNames: ["openai_api_key"],
+    apiKeyNames: ["OPENAI_API_KEY"],
     url: "https://platform.openai.com/docs/guides/codex",
   },
   cursor: {
     displayName: "Cursor CLI",
-    apiKeyNames: ["cursor_api_key"],
+    apiKeyNames: ["CURSOR_API_KEY"],
     url: "https://cursor.com/",
   },
   gemini: {
     displayName: "Gemini CLI",
-    apiKeyNames: ["google_api_key", "gemini_api_key"],
+    apiKeyNames: ["GOOGLE_API_KEY", "GEMINI_API_KEY"],
     url: "https://ai.google.dev/gemini-api/docs",
   },
   opencode: {
@@ -51,6 +51,10 @@ export const AgentName = type.enumerated(...Object.keys(agentsManifest));
 
 export type AgentApiKeyName = (typeof agentsManifest)[AgentName]["apiKeyNames"][number];
 
+// effort level type - controls model selection and thinking level
+export const Effort = type.enumerated("nothink", "think", "max");
+export type Effort = typeof Effort.infer;
+
 // base interface for common payload event fields
 interface BasePayloadEvent {
   issue_number?: number;
@@ -279,6 +283,12 @@ export interface Payload extends DispatchOptions {
    */
   modes: readonly Mode[];
 
+  /**
+   * Effort level for model selection (nothink, think, max)
+   * Defaults to "think" if not specified
+   */
+  readonly effort?: Effort;
+
   /**
    * Optional IDs of the issue, PR, or comment that the agent is working on
    */

fixtures/cursor-effort.ts

@@ -0,0 +1,30 @@
+import type { Effort, Payload } from "../external.ts";
+
+/**
+ * Test fixture for Cursor effort levels.
+ * Runs all three effort levels in sequence.
+ *
+ * Run with:
+ *   AGENT_OVERRIDE=cursor pnpm play cursor-effort.ts
+ *
+ * Effort levels:
+ *   - "nothink": auto (default model)
+ *   - "think": auto (default model)
+ *   - "max": opus-4.5-thinking
+ *
+ * Note: If project has .cursor/cli.json with "model" specified,
+ * that takes precedence over effort-based model selection.
+ */
+
+const efforts: Effort[] = ["nothink", "think", "max"];
+
+export default efforts.map((effort) => ({
+  "~pullfrog": true,
+  agent: "cursor",
+  prompt: "What is 2 + 2? Reply with just the number.",
+  event: {
+    trigger: "workflow_dispatch",
+  },
+  modes: [],
+  effort,
+})) satisfies Payload[];

fixtures/gemini-effort.ts

@@ -0,0 +1,27 @@
+import type { Effort, Payload } from "../external.ts";
+
+/**
+ * Test fixture for Gemini effort levels.
+ * Runs all three effort levels in sequence.
+ *
+ * Run with:
+ *   AGENT_OVERRIDE=gemini pnpm play gemini-effort.ts
+ *
+ * Effort levels:
+ *   - "nothink": gemini-2.5-flash + LOW thinking
+ *   - "think": gemini-2.5-flash + HIGH thinking
+ *   - "max": gemini-2.5-pro + HIGH thinking
+ */
+
+const efforts: Effort[] = ["nothink", "think", "max"];
+
+export default efforts.map((effort) => ({
+  "~pullfrog": true,
+  agent: "gemini",
+  prompt: "What is 2 + 2? Reply with just the number.",
+  event: {
+    trigger: "workflow_dispatch",
+  },
+  modes: [],
+  effort,
+})) satisfies Payload[];

get-installation-token/action.yml

@@ -0,0 +1,21 @@
+name: "Get Installation Token"
+description: "Get a GitHub App installation token for the current repository"
+author: "Pullfrog"
+
+inputs:
+  repos:
+    description: "Comma-separated list of additional repo names to grant access to (e.g., 'repo1,repo2'). Current repo is always included."
+    required: false
+
+outputs:
+  token:
+    description: "GitHub App installation token"
+
+runs:
+  using: "node24"
+  main: "entry"
+  post: "entry"
+
+branding:
+  icon: "key"
+  color: "green"

get-installation-token/entry.ts

@@ -0,0 +1,69 @@
+#!/usr/bin/env node
+
+/**
+ * entry point for get-installation-token action.
+ * handles both main and post execution using the isPost state pattern.
+ */
+
+import * as core from "@actions/core";
+import { acquireInstallationToken, revokeInstallationToken } from "./token.ts";
+
+const STATE_TOKEN = "token";
+const STATE_IS_POST = "isPost";
+
+async function main(): Promise<void> {
+  core.saveState(STATE_IS_POST, "true");
+
+  const reposInput = core.getInput("repos");
+  const additionalRepos = reposInput
+    ? reposInput
+        .split(",")
+        .map((r) => r.trim())
+        .filter(Boolean)
+    : [];
+
+  const token = await acquireInstallationToken({ repos: additionalRepos });
+
+  // mask the token in logs
+  core.setSecret(token);
+
+  // save token to state for post cleanup
+  core.saveState(STATE_TOKEN, token);
+
+  // set as output
+  core.setOutput("token", token);
+
+  const scope = additionalRepos.length
+    ? `current repo + ${additionalRepos.join(", ")}`
+    : "current repo only";
+  core.info(`» installation token acquired (${scope})`);
+}
+
+async function post(): Promise<void> {
+  const token = core.getState(STATE_TOKEN);
+
+  if (!token) {
+    core.debug("no token found in state, skipping revocation");
+    return;
+  }
+
+  await revokeInstallationToken(token);
+  core.info("» installation token revoked");
+}
+
+async function run(): Promise<void> {
+  try {
+    const isPost = core.getState(STATE_IS_POST) === "true";
+
+    if (isPost) {
+      await post();
+    } else {
+      await main();
+    }
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    core.setFailed(message);
+  }
+}
+
+await run();

get-installation-token/token.ts

@@ -0,0 +1,14 @@
+/**
+ * token acquisition and revocation for get-installation-token action.
+ * reuses the existing github.ts utilities.
+ */
+
+import { acquireNewToken, revokeGitHubInstallationToken } from "../utils/github.ts";
+
+export async function acquireInstallationToken(opts?: { repos?: string[] }): Promise<string> {
+  return acquireNewToken(opts);
+}
+
+export async function revokeInstallationToken(token: string): Promise<void> {
+  return revokeGitHubInstallationToken(token);
+}

main.ts

@@ -7,8 +7,7 @@ import { encode as toonEncode } from "@toon-format/toon";
 import { type } from "arktype";
 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";
+import { type AgentName, agentsManifest, Effort, type Payload } from "./external.ts";
 import { ensureProgressCommentUpdated, reportProgress } from "./mcp/comment.ts";
 import { createMcpConfigs } from "./mcp/config.ts";
 import { startMcpHttpServer } from "./mcp/server.ts";
@@ -18,29 +17,13 @@ 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";
-import {
-  createOctokit,
-  parseRepoContext,
-  setupGitHubInstallationToken,
-} from "./utils/github.ts";
+import { createOctokit, parseRepoContext, setupGitHubInstallationToken } from "./utils/github.ts";
 import { setupGitAuth, setupGitConfig } from "./utils/setup.ts";
 import { Timer } from "./utils/timer.ts";
 
-// runtime validation using agents (needed for ArkType)
-// Note: The AgentName type is defined in external.ts, this is the runtime validator
-
-export const AgentInputKey = type.enumerated(
-  ...Object.values(agents).flatMap((agent) => agent.apiKeyNames)
-);
-export type AgentInputKey = typeof AgentInputKey.infer;
-
-const keyInputDefs = flatMorph(agents, (_, agent) =>
-  agent.apiKeyNames.map((inputKey) => [inputKey, "string | undefined?"] as const)
-);
-
 export const Inputs = type({
   prompt: "string",
-  ...keyInputDefs,
+  "effort?": Effort,
 });
 
 export type Inputs = typeof Inputs.infer;
@@ -84,7 +67,6 @@ export async function main(inputs: Inputs): Promise<MainResult> {
 
     // phase 3: resolve agent (needs repo settings)
     const agent = resolveAgent({
-      inputs,
       payload,
       repoSettings: githubSetup.repoSettings,
     });
@@ -93,7 +75,6 @@ export async function main(inputs: Inputs): Promise<MainResult> {
     // 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,
     });
@@ -225,24 +206,19 @@ export async function main(inputs: Inputs): Promise<MainResult> {
 }
 
 /**
- * Get agents that have matching API keys in the inputs
+ * Check if an agent has API keys available (from process.env)
  */
-/**
- * Check if an agent has API keys available (inputs or process.env for opencode)
- */
-function agentHasApiKeys(agent: Agent, inputs: Inputs): boolean {
+function agentHasApiKeys(agent: Agent): 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;
+    // opencode accepts any API_KEY from environment
     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]);
+  // check if any of the agent's expected keys are in environment
+  return agent.apiKeyNames.some((envKey) => !!process.env[envKey]);
 }
 
-function getAvailableAgents(inputs: Inputs): Agent[] {
-  return Object.values(agents).filter((agent) => agentHasApiKeys(agent, inputs));
+function getAvailableAgents(): Agent[] {
+  return Object.values(agents).filter((agent) => agentHasApiKeys(agent));
 }
 
 /**
@@ -275,7 +251,7 @@ function buildMissingApiKeyError(params: { agent: Agent; owner: string; name: st
   } else {
     const inputKeys =
       params.agent.apiKeyNames.length > 0 ? params.agent.apiKeyNames : getAllPossibleKeyNames();
-    const secretNames = inputKeys.map((key) => `\`${key.toUpperCase()}\``);
+    const secretNames = inputKeys.map((key) => `\`${key}\``);
     secretNameList = inputKeys.length === 1 ? secretNames[0] : `one of ${secretNames.join(" or ")}`;
   }
 
@@ -361,16 +337,16 @@ async function initializeGitHub(token: string): Promise<GitHubSetup> {
 }
 
 function resolveAgent({
-  inputs,
   payload,
   repoSettings,
 }: {
-  inputs: Inputs;
   payload: Payload;
   repoSettings: RepoSettings;
 }): Agent {
   const agentOverride = process.env.AGENT_OVERRIDE as AgentName | undefined;
-  log.debug(`» determineAgent: agentOverride=${agentOverride}, payload.agent=${payload.agent}, repoSettings.defaultAgent=${repoSettings.defaultAgent}`);
+  log.debug(
+    `» determineAgent: agentOverride=${agentOverride}, payload.agent=${payload.agent}, repoSettings.defaultAgent=${repoSettings.defaultAgent}`
+  );
   const configuredAgentName = agentOverride || payload.agent || repoSettings.defaultAgent || null;
 
   if (configuredAgentName) {
@@ -388,13 +364,13 @@ function resolveAgent({
     }
 
     // for repo-level defaults, check if agent has matching keys before selecting
-    if (agentHasApiKeys(agent, inputs)) {
+    if (agentHasApiKeys(agent)) {
       log.info(`Selected configured agent: ${agent.name}`);
       return agent;
     }
 
     // fall through to auto-selection
-    const availableAgents = getAvailableAgents(inputs);
+    const availableAgents = getAvailableAgents();
     log.warning(
       `Repo default agent ${agent.name} has no matching API keys. Available: ${
         availableAgents.map((a) => a.name).join(", ") || "none"
@@ -402,7 +378,7 @@ function resolveAgent({
     );
   }
 
-  const availableAgents = getAvailableAgents(inputs);
+  const availableAgents = getAvailableAgents();
   if (availableAgents.length === 0) {
     throw new Error("no agents available - missing API keys");
   }
@@ -425,8 +401,13 @@ function parsePayload(inputs: Inputs): Payload {
     if (!("~pullfrog" in parsedPrompt)) {
       throw new Error();
     }
-    return parsedPrompt as Payload;
+    // internal invocation: use effort from payload, fallback to input, default to "think"
+    return {
+      ...parsedPrompt,
+      effort: parsedPrompt.effort ?? inputs.effort ?? "think",
+    } as Payload;
   } catch {
+    // external invocation: use effort from input
     return {
       "~pullfrog": true,
       agent: null,
@@ -435,6 +416,7 @@ function parsePayload(inputs: Inputs): Payload {
         trigger: "unknown",
       },
       modes,
+      effort: inputs.effort ?? "think",
     };
   }
 }
@@ -447,22 +429,22 @@ async function installAgentCli(params: { agent: Agent; token: string }): Promise
   return params.agent.install();
 }
 
-function collectApiKeys(agent: Agent, inputs: Inputs): Record<string, string> {
+function collectApiKeys(agent: Agent): Record<string, string> {
   const apiKeys: Record<string, string> = {};
-  const inputsRecord = inputs as Record<string, string | undefined>;
 
-  for (const inputKey of agent.apiKeyNames) {
-    const value = inputsRecord[inputKey];
+  // read API keys from environment variables
+  for (const envKey of agent.apiKeyNames) {
+    const value = process.env[envKey];
     if (value) {
-      apiKeys[inputKey] = value;
+      apiKeys[envKey] = value;
     }
   }
 
-  // for OpenCode: also check process.env for any API_KEY variables
+  // for OpenCode: 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")) {
-        apiKeys[key.toLowerCase()] = value;
+        apiKeys[key] = value;
       }
     }
   }
@@ -470,13 +452,8 @@ function collectApiKeys(agent: Agent, inputs: Inputs): Record<string, string> {
   return apiKeys;
 }
 
-function validateApiKey(params: {
-  agent: Agent;
-  inputs: Inputs;
-  owner: string;
-  name: string;
-}): ApiKeySetup {
-  const apiKeys = collectApiKeys(params.agent, params.inputs);
+function validateApiKey(params: { agent: Agent; owner: string; name: string }): ApiKeySetup {
+  const apiKeys = collectApiKeys(params.agent);
 
   if (Object.keys(apiKeys).length === 0) {
     return {
@@ -497,7 +474,8 @@ function validateApiKey(params: {
 }
 
 async function runAgent(ctx: AgentContext): Promise<AgentResult> {
-  log.info(`Running ${ctx.agent.name}...`);
+  const effort = ctx.payload.effort ?? "think";
+  log.info(`Running ${ctx.agent.name} with effort=${effort}...`);
   // strip context from event
   const { context: _context, ...eventWithoutContext } = ctx.payload.event;
   // format: prompt + two newlines + TOON encoded event
@@ -516,6 +494,7 @@ async function runAgent(ctx: AgentContext): Promise<AgentResult> {
       defaultBranch: ctx.repo.default_branch,
       isPublic: !ctx.repo.private,
     },
+    effort,
   });
 }
 

mcp/comment.ts

@@ -1,10 +1,16 @@
+import * as core from "@actions/core";
 import { type } from "arktype";
 import type { Payload } from "../external.ts";
 import { agentsManifest } from "../external.ts";
 import type { ToolContext } from "../main.ts";
 import { fetchWorkflowRunInfo } from "../utils/api.ts";
 import { buildPullfrogFooter, stripExistingFooter } from "../utils/buildPullfrogFooter.ts";
-import { createOctokit, getGitHubInstallationToken, parseRepoContext, type OctokitWithPlugins } from "../utils/github.ts";
+import {
+  createOctokit,
+  getGitHubInstallationToken,
+  type OctokitWithPlugins,
+  parseRepoContext,
+} from "../utils/github.ts";
 import { execute, tool } from "./shared.ts";
 
 /**
@@ -14,6 +20,8 @@ import { execute, tool } from "./shared.ts";
  */
 export const LEAPING_INTO_ACTION_PREFIX = "Leaping into action";
 
+const isGitHubActions = !!process.env.GITHUB_ACTIONS;
+
 interface BuildCommentFooterParams {
   payload: Payload;
   octokit?: OctokitWithPlugins | undefined;
@@ -79,7 +87,11 @@ function buildImplementPlanLink(
   return `[Implement plan ➔](${apiUrl}/trigger/${owner}/${repo}/${issueNumber}?action=implement&comment_id=${commentId})`;
 }
 
-async function addFooter(body: string, payload: Payload, octokit?: OctokitWithPlugins): Promise<string> {
+async function addFooter(
+  body: string,
+  payload: Payload,
+  octokit?: OctokitWithPlugins
+): Promise<string> {
   const bodyWithoutFooter = stripExistingFooter(body);
   const footer = await buildCommentFooter({ payload, octokit });
   return `${bodyWithoutFooter}${footer}`;
@@ -188,6 +200,10 @@ export const ReportProgress = type({
   body: type.string.describe("the progress update content to share"),
 });
 
+
+/** Updates job summary with the given text if running in GitHub Actions. */
+const updateSummary = (text: string) => isGitHubActions && core.summary.addRaw(text).write({ overwrite: true });
+
 /**
  * Standalone function to report progress to GitHub comment.
  * Can be called directly without going through the MCP tool interface.
@@ -234,6 +250,8 @@ export async function reportProgress(
 
     progressCommentWasUpdated = true;
 
+    await updateSummary(bodyWithFooter);
+
     return {
       commentId: result.data.id,
       url: result.data.html_url,
@@ -281,6 +299,8 @@ export async function reportProgress(
       body: bodyWithPlanLink,
     });
 
+    await updateSummary(bodyWithPlanLink);
+
     return {
       commentId: updateResult.data.id,
       url: updateResult.data.html_url,
@@ -289,6 +309,8 @@ export async function reportProgress(
     };
   }
 
+  await updateSummary(initialBody);
+
   return {
     commentId: result.data.id,
     url: result.data.html_url,
@@ -481,6 +503,6 @@ export function ReplyToReviewCommentTool(ctx: ToolContext) {
         body: result.data.body,
         in_reply_to_id: result.data.in_reply_to_id,
       };
-    }),
+    }, "reply_to_review_comment"),
   });
 }

mcp/reviewComments.ts

@@ -93,95 +93,95 @@ export function GetReviewCommentsTool(ctx: ToolContext) {
       "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(async ({ pull_number, review_id }) => {
-      // fetch all review threads using graphql
-      const response = await ctx.octokit.graphql<GraphQLResponse>(REVIEW_THREADS_QUERY, {
-        owner: ctx.owner,
-        repo: ctx.name,
-        pullNumber: pull_number,
-      });
+        // fetch all review threads using graphql
+        const response = await ctx.octokit.graphql<GraphQLResponse>(REVIEW_THREADS_QUERY, {
+          owner: ctx.owner,
+          repo: ctx.name,
+          pullNumber: pull_number,
+        });
 
-      const pullRequest = response.repository?.pullRequest;
-      if (!pullRequest) {
-        return {
-          review_id,
-          pull_number,
-          comments: [],
-          count: 0,
-        };
-      }
-
-      const threadNodes = pullRequest.reviewThreads?.nodes;
-      if (!threadNodes) {
-        return {
-          review_id,
-          pull_number,
-          comments: [],
-          count: 0,
-        };
-      }
-
-      const allComments: {
-        id: number;
-        body: string;
-        path: string;
-        line: number | null;
-        side: "LEFT" | "RIGHT";
-        start_line: number | null;
-        start_side: "LEFT" | "RIGHT" | null;
-        user: string | null;
-        created_at: string;
-        updated_at: string;
-        html_url: string;
-        in_reply_to_id: number | null;
-        pull_request_review_id: number | null;
-      }[] = [];
-
-      // iterate through all threads (filter out nulls)
-      for (const thread of threadNodes) {
-        if (!thread?.comments?.nodes) continue;
-
-        // filter out null comments
-        const threadComments = thread.comments.nodes.filter(
-          (c): c is GraphQLReviewComment => c !== null
-        );
-        if (threadComments.length === 0) continue;
-
-        // find the root comment (the one with replyTo == null) to determine thread ownership
-        const rootComment = threadComments.find((c) => c.replyTo === null);
-        if (!rootComment) continue;
-
-        // check if this thread belongs to the target review using the root comment
-        const threadBelongsToReview = rootComment.pullRequestReview?.databaseId === review_id;
-        if (!threadBelongsToReview) continue;
-
-        // include all comments from this thread (original + replies)
-        // side info comes from thread level, not comment level
-        for (const comment of threadComments) {
-          allComments.push({
-            id: comment.databaseId,
-            body: comment.body,
-            path: comment.path,
-            line: comment.line,
-            start_line: comment.startLine,
-            side: thread.diffSide,
-            start_side: thread.startDiffSide,
-            user: comment.author?.login ?? null,
-            created_at: comment.createdAt,
-            updated_at: comment.updatedAt,
-            html_url: comment.url,
-            in_reply_to_id: comment.replyTo?.databaseId ?? null,
-            pull_request_review_id: comment.pullRequestReview?.databaseId ?? null,
-          });
+        const pullRequest = response.repository?.pullRequest;
+        if (!pullRequest) {
+          return {
+            review_id,
+            pull_number,
+            comments: [],
+            count: 0,
+          };
         }
-      }
 
-      return {
-        review_id,
-        pull_number,
-        comments: allComments,
-        count: allComments.length,
-      };
-    }),
+        const threadNodes = pullRequest.reviewThreads?.nodes;
+        if (!threadNodes) {
+          return {
+            review_id,
+            pull_number,
+            comments: [],
+            count: 0,
+          };
+        }
+
+        const allComments: {
+          id: number;
+          body: string;
+          path: string;
+          line: number | null;
+          side: "LEFT" | "RIGHT";
+          start_line: number | null;
+          start_side: "LEFT" | "RIGHT" | null;
+          user: string | null;
+          created_at: string;
+          updated_at: string;
+          html_url: string;
+          in_reply_to_id: number | null;
+          pull_request_review_id: number | null;
+        }[] = [];
+
+        // iterate through all threads (filter out nulls)
+        for (const thread of threadNodes) {
+          if (!thread?.comments?.nodes) continue;
+
+          // filter out null comments
+          const threadComments = thread.comments.nodes.filter(
+            (c): c is GraphQLReviewComment => c !== null
+          );
+          if (threadComments.length === 0) continue;
+
+          // find the root comment (the one with replyTo == null) to determine thread ownership
+          const rootComment = threadComments.find((c) => c.replyTo === null);
+          if (!rootComment) continue;
+
+          // check if this thread belongs to the target review using the root comment
+          const threadBelongsToReview = rootComment.pullRequestReview?.databaseId === review_id;
+          if (!threadBelongsToReview) continue;
+
+          // include all comments from this thread (original + replies)
+          // side info comes from thread level, not comment level
+          for (const comment of threadComments) {
+            allComments.push({
+              id: comment.databaseId,
+              body: comment.body,
+              path: comment.path,
+              line: comment.line,
+              start_line: comment.startLine,
+              side: thread.diffSide,
+              start_side: thread.startDiffSide,
+              user: comment.author?.login ?? null,
+              created_at: comment.createdAt,
+              updated_at: comment.updatedAt,
+              html_url: comment.url,
+              in_reply_to_id: comment.replyTo?.databaseId ?? null,
+              pull_request_review_id: comment.pullRequestReview?.databaseId ?? null,
+            });
+          }
+        }
+
+        return {
+          review_id,
+          pull_number,
+          comments: allComments,
+          count: allComments.length,
+        };
+      }),
   });
 }
 

mcp/shared.ts

@@ -2,6 +2,7 @@ import type { StandardSchemaV1 } from "@standard-schema/spec";
 import { encode as toonEncode } from "@toon-format/toon";
 import type { FastMCP, Tool } from "fastmcp";
 import type { ToolContext } from "../main.ts";
+import { formatJsonValue, log } from "../utils/cli.ts";
 
 export const tool = <const params>(toolDef: Tool<any, StandardSchemaV1<params>>) => toolDef;
 
@@ -36,13 +37,22 @@ 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.
+ * @param fn - the function to execute
+ * @param toolName - optional tool name for error logging
  */
-export const execute = <T>(fn: (params: T) => Promise<Record<string, any> | string>) => {
+export const execute = <T>(
+  fn: (params: T) => Promise<Record<string, any> | string>,
+  toolName?: string
+) => {
   return async (params: T): Promise<ToolResult> => {
     try {
       const result = await fn(params);
       return handleToolSuccess(result);
     } catch (error) {
+      const errorMessage = error instanceof Error ? error.message : String(error);
+      const prefix = toolName ? `[${toolName}]` : "tool";
+      log.error(`${prefix} error: ${errorMessage}`);
+      log.debug(`${prefix} params: ${formatJsonValue(params)}`);
       return handleToolError(error);
     }
   };

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pullfrog/action",
-  "version": "0.0.156",
+  "version": "0.0.157",
   "type": "module",
   "files": [
     "index.js",
@@ -13,7 +13,7 @@
     "main.d.ts"
   ],
   "scripts": {
-    "test": "echo \"Error: no test specified\" && exit 1",
+    "test": "vitest",
     "typecheck": "tsc --noEmit",
     "build": "node esbuild.config.js",
     "play": "node play.ts",
@@ -47,7 +47,8 @@
     "arg": "^5.0.2",
     "esbuild": "^0.25.9",
     "husky": "^9.0.0",
-    "typescript": "^5.9.3"
+    "typescript": "^5.9.3",
+    "vitest": "^4.0.17"
   },
   "repository": {
     "type": "git",

play.ts

@@ -3,10 +3,8 @@ import { existsSync, readFileSync } from "node:fs";
 import { extname, join, resolve } from "node:path";
 import { pathToFileURL } from "node:url";
 import { fromHere } from "@ark/fs";
-import { flatMorph } from "@ark/util";
 import arg from "arg";
 import { config } from "dotenv";
-import { agents } from "./agents/index.ts";
 import type { AgentResult } from "./agents/shared.ts";
 import { type Inputs, main } from "./main.ts";
 import { log } from "./utils/cli.ts";
@@ -25,26 +23,9 @@ export async function run(prompt: string): Promise<AgentResult> {
     const originalCwd = process.cwd();
     process.chdir(tempDir);
 
-    // check if prompt is a pullfrog payload and extract agent
-    // note: agent from payload will be used by determineAgent with highest precedence
-    // we don't need to extract it here since main() will parse the payload
-    const inputs = {
+    const inputs: Inputs = {
       prompt,
-      ...flatMorph(agents, (_, agent) => {
-        // for OpenCode, scan all API_KEY environment variables
-        if (agent.name === "opencode") {
-          const opencodeKeys: Array<[string, string | undefined]> = [];
-          for (const [key, value] of Object.entries(process.env)) {
-            if (value && typeof value === "string" && key.includes("API_KEY")) {
-              opencodeKeys.push([key.toLowerCase(), value]);
-            }
-          }
-          return opencodeKeys;
-        }
-        // for other agents, use apiKeyNames
-        return agent.apiKeyNames.map((inputKey) => [inputKey, process.env[inputKey.toUpperCase()]]);
-      }),
-    } as Required<Inputs>;
+    };
 
     const result = await main(inputs);
 
@@ -205,6 +186,30 @@ Examples:
 
         if (typeof module.default === "string") {
           prompt = module.default;
+        } else if (Array.isArray(module.default)) {
+          // Array of Payloads - run each in sequence
+          const payloads = module.default;
+          log.info(`Running ${payloads.length} payloads in sequence...`);
+
+          let allSuccess = true;
+          for (let i = 0; i < payloads.length; i++) {
+            const payload = payloads[i];
+            const label = payload.effort
+              ? `[${i + 1}/${payloads.length}] effort=${payload.effort}`
+              : `[${i + 1}/${payloads.length}]`;
+            log.info(`\n${"=".repeat(60)}`);
+            log.info(`${label}`);
+            log.info(`${"=".repeat(60)}\n`);
+
+            const payloadPrompt = JSON.stringify(payload, null, 2);
+            const result = await run(payloadPrompt);
+            if (!result.success) {
+              allSuccess = false;
+              log.error(`Payload ${i + 1} failed`);
+            }
+          }
+
+          process.exit(allSuccess ? 0 : 1);
         } else if (typeof module.default === "object") {
           // Payload objects (with ~pullfrog) should be stringified
           prompt = JSON.stringify(module.default, null, 2);

tsconfig.json

@@ -5,6 +5,7 @@
     "target": "ESNext",
     "moduleResolution": "NodeNext",
     "lib": ["ESNext"],
+    "types": ["vitest/globals"],
     "allowImportingTsExtensions": true,
     "rewriteRelativeImportExtensions": true,
     "skipLibCheck": true,

utils/cli.ts

@@ -178,7 +178,10 @@ async function summaryTable(
   if (title) {
     core.info(`\n${title}`);
   }
-  const tableText = formattedRows.map((row) => row.map((cell) => cell.data).join(" | ")).join("\n");
+  const tableData = formattedRows.map((row) => row.map((cell) => cell.data));
+  const tableText = isGitHubActions
+    ? tableData.map((row) => row.join(" | ")).join("\n")
+    : table(tableData);
   core.info(`\n${tableText}\n`);
 }
 

utils/github.ts

@@ -54,12 +54,17 @@ function isOIDCAvailable(): boolean {
   );
 }
 
-async function acquireTokenViaOIDC(): Promise<string> {
+async function acquireTokenViaOIDC(opts?: { repos?: string[] }): Promise<string> {
   log.info("» generating OIDC token...");
 
   const oidcToken = await core.getIDToken("pullfrog-api");
 
   const apiUrl = process.env.API_URL || "https://pullfrog.com";
+  const params = new URLSearchParams();
+  if (opts?.repos?.length) {
+    params.set("repos", opts.repos.join(","));
+  }
+  const queryString = params.toString() ? `?${params.toString()}` : "";
 
   log.info("» exchanging OIDC token for installation token...");
 
@@ -68,7 +73,7 @@ async function acquireTokenViaOIDC(): Promise<string> {
   const timeoutId = setTimeout(() => controller.abort(), timeoutMs);
 
   try {
-    const tokenResponse = await fetch(`${apiUrl}/api/github/installation-token`, {
+    const tokenResponse = await fetch(`${apiUrl}/api/github/installation-token${queryString}`, {
       method: "POST",
       headers: {
         Authorization: `Bearer ${oidcToken}`,
@@ -84,7 +89,11 @@ async function acquireTokenViaOIDC(): Promise<string> {
     }
 
     const tokenData = (await tokenResponse.json()) as InstallationToken;
-    log.info(`» installation token obtained for ${tokenData.repository || "all repositories"}`);
+    const owner = tokenData.repository?.split("/")[0];
+    const repoList = opts?.repos?.length
+      ? [tokenData.repository, ...opts.repos.map((r) => `${owner}/${r}`)].join(", ")
+      : tokenData.repository;
+    log.info(`» installation token obtained for ${repoList}`);
 
     return tokenData.token;
   } catch (error) {
@@ -238,9 +247,9 @@ async function acquireTokenViaGitHubApp(): Promise<string> {
   return token;
 }
 
-async function acquireNewToken(): Promise<string> {
+export async function acquireNewToken(opts?: { repos?: string[] }): Promise<string> {
   if (isOIDCAvailable()) {
-    return await retry(() => acquireTokenViaOIDC(), { label: "token exchange" });
+    return await retry(() => acquireTokenViaOIDC(opts), { label: "token exchange" });
   } else {
     return await acquireTokenViaGitHubApp();
   }
@@ -277,7 +286,7 @@ export function getGitHubInstallationToken(): string {
   return githubInstallationToken;
 }
 
-async function revokeGitHubInstallationToken(token: string): Promise<void> {
+export async function revokeGitHubInstallationToken(token: string): Promise<void> {
   const apiUrl = process.env.GITHUB_API_URL || "https://api.github.com";
 
   try {

utils/timer.test.ts

@@ -0,0 +1,110 @@
+import { Timer } from './timer.ts';
+import * as cli from './cli.ts';
+
+describe('Timer', () => {
+  beforeEach(() => {
+    vi.spyOn(cli.log, 'debug');
+    // Mock Date.now to have predictable timestamps
+    vi.useFakeTimers();
+  });
+
+  afterEach(() => {
+    vi.restoreAllMocks();
+    vi.useRealTimers();
+  });
+
+  describe('constructor', () => {
+    it('should initialize with current timestamp', () => {
+      const mockTime = 1000000;
+      vi.setSystemTime(mockTime);
+
+      const timer = new Timer();
+      timer.checkpoint('test');
+
+      expect(cli.log.debug).toHaveBeenCalledWith(
+        expect.stringContaining('test')
+      );
+    });
+  });
+
+  describe('checkpoint', () => {
+    it('should log duration from initial timestamp on first checkpoint', () => {
+      const startTime = 1000000;
+      vi.setSystemTime(startTime);
+      const timer = new Timer();
+
+      const checkpointTime = startTime + 100;
+      vi.setSystemTime(checkpointTime);
+      timer.checkpoint('first');
+
+      expect(cli.log.debug).toHaveBeenCalledWith('» first: 100ms');
+    });
+
+    it('should log duration from last checkpoint on subsequent checkpoints', () => {
+      const startTime = 1000000;
+      vi.setSystemTime(startTime);
+      const timer = new Timer();
+
+      // First checkpoint
+      const firstCheckpointTime = startTime + 50;
+      vi.setSystemTime(firstCheckpointTime);
+      timer.checkpoint('first');
+
+      // Second checkpoint
+      const secondCheckpointTime = firstCheckpointTime + 75;
+      vi.setSystemTime(secondCheckpointTime);
+      timer.checkpoint('second');
+
+      expect(cli.log.debug).toHaveBeenCalledTimes(2);
+      expect(cli.log.debug).toHaveBeenNthCalledWith(1, '» first: 50ms');
+      expect(cli.log.debug).toHaveBeenNthCalledWith(2, '» second: 75ms');
+    });
+
+    it('should handle multiple checkpoints correctly', () => {
+      const startTime = 1000000;
+      vi.setSystemTime(startTime);
+      const timer = new Timer();
+
+      // First checkpoint
+      vi.setSystemTime(startTime + 10);
+      timer.checkpoint('step1');
+
+      // Second checkpoint
+      vi.setSystemTime(startTime + 25);
+      timer.checkpoint('step2');
+
+      // Third checkpoint
+      vi.setSystemTime(startTime + 45);
+      timer.checkpoint('step3');
+
+      expect(cli.log.debug).toHaveBeenCalledTimes(3);
+      expect(cli.log.debug).toHaveBeenNthCalledWith(1, '» step1: 10ms');
+      expect(cli.log.debug).toHaveBeenNthCalledWith(2, '» step2: 15ms');
+      expect(cli.log.debug).toHaveBeenNthCalledWith(3, '» step3: 20ms');
+    });
+
+    it('should handle zero duration correctly', () => {
+      const startTime = 1000000;
+      vi.setSystemTime(startTime);
+      const timer = new Timer();
+
+      // Checkpoint immediately
+      timer.checkpoint('immediate');
+
+      expect(cli.log.debug).toHaveBeenCalledWith('» immediate: 0ms');
+    });
+
+    it('should handle custom checkpoint names', () => {
+      const startTime = 1000000;
+      vi.setSystemTime(startTime);
+      const timer = new Timer();
+
+      vi.setSystemTime(startTime + 200);
+      timer.checkpoint('Custom Checkpoint Name');
+
+      expect(cli.log.debug).toHaveBeenCalledWith(
+        '» Custom Checkpoint Name: 200ms'
+      );
+    });
+  });
+});

vitest.config.ts

@@ -0,0 +1,9 @@
+import { defineConfig } from 'vitest/config';
+
+export default defineConfig({
+  test: {
+    globals: true,
+    environment: 'node',
+    exclude: ['node_modules', '.temp'],
+  },
+});