Files
shockbot/agents/claude.ts
T
Colin McDonnell b6658ddbc1 improve agent CI matrix, token permissions, and waitlist follower backfill (#313)
* add workflows permission to git token and waitlist improvements

- add `workflows` to `InstallationTokenPermissions` type in both action and API token routes
- include `workflows: write` in the git token so agents can push workflow file changes
- add `githubFollowers` field to WaitlistSignup schema with migration
- add script to populate waitlist followers from GitHub API
- add frog-green-square-border logo asset

Co-authored-by: Cursor <cursoragent@cursor.com>

* improve CI, agent logging, token permissions, and delegation guardrails

- add format check and build step to root CI job
- standardize agent model/effort log lines across all agents
- fix GitHub App permissions types to match OpenAPI schema (workflows is write-only)
- improve delegation error message to prevent subagent recursion
- demote noisy OpenCode stderr to debug level
- add subagent delegation rules to resolved instructions

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix graphql partial error handling, update delegation message, add workflow_run fixtures

Co-authored-by: Cursor <cursoragent@cursor.com>

* remove module-level env var throws that break CI build

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix logging bug and type hole from PR review

- use batch-local notFound counter so per-batch log doesn't undercount
- add workflows to WorkflowTokenPermissions so wire type matches what action sends

Co-authored-by: Cursor <cursoragent@cursor.com>

* lazy-init appOctokit to fix next build without env vars

Co-authored-by: Cursor <cursoragent@cursor.com>

* drop pnpm build from CI test workflow

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix delegate-effort test regex to match actual log format, disable fail-fast for agnostic tests

the test was matching `running \w+ with effort=auto` but the actual log
line from shared.ts is `» effort:  auto`. also temporarily set
fail-fast: false on action-agnostic so all failures surface at once.

Co-authored-by: Cursor <cursoragent@cursor.com>

* disable fail-fast in action workflow too, relax ci.test.ts to match

both workflow files now use fail-fast: false for agnostic tests so all
matrix jobs run to completion. the ci consistency test now checks that
the two workflows agree rather than requiring true.

Co-authored-by: Cursor <cursoragent@cursor.com>

* restore fail-fast: true now that all agnostic tests pass

Co-authored-by: Cursor <cursoragent@cursor.com>

* skip agent tests in CI when agent harness file didn't change

adds action/test/changed-agents.sh which reads the PR diff (via
dorny/paths-filter) and outputs only agents whose harness file was
modified. the action-agents matrix now uses this dynamic list instead
of a hardcoded array, so e.g. a PR touching only cursor.ts runs 6
jobs instead of 30.

Co-authored-by: Cursor <cursoragent@cursor.com>

* update ci.test.ts to validate dynamic agent matrix

the test now checks that the matrix references the changes job output
and that changed-agents.sh correctly discovers all agents.

Co-authored-by: Cursor <cursoragent@cursor.com>

* parallelize action jobs and use claude canary fallback for shared changes

runs action-agents in parallel with action-agnostic after root/changes, and updates changed-agents logic so shared or non-harness action runtime changes run only claude while harness-specific edits run only those changed agents.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-02-16 22:36:29 +00:00

317 lines
10 KiB
TypeScript

// changes to effort level configuration should be reflected in wiki/effort.md and docs/effort.mdx
// changes to tool permissions should be reflected in wiki/granular-tools.md
// changes to web search configuration should be reflected in wiki/websearch.md
import { mkdirSync, writeFileSync } from "node:fs";
import { join } from "node:path";
import type { SDKMessage } from "@anthropic-ai/claude-agent-sdk";
import type { Effort } from "../external.ts";
import { ghPullfrogMcpName } from "../external.ts";
import packageJson from "../package.json" with { type: "json" };
import { markActivity } from "../utils/activity.ts";
import { log } from "../utils/cli.ts";
import { installFromNpmTarball } from "../utils/install.ts";
import { spawn } from "../utils/subprocess.ts";
import { ThinkingTimer } from "../utils/timer.ts";
import { type AgentRunContext, agent } from "./shared.ts";
// model selection based on effort level
// these are aliases that always resolve to the latest version
const claudeEffortModels: Record<Effort, string> = {
mini: "sonnet",
auto: "opus",
max: "opus",
};
// Claude Code CLI --effort level per pullfrog effort
// null = use default (high). "max" is Opus 4.6 only.
const claudeEffortLevels: Record<Effort, string | null> = {
mini: null,
auto: null,
max: "max",
};
/**
* Build disallowedTools list from payload permissions.
*/
function buildDisallowedTools(ctx: AgentRunContext): string[] {
const disallowed: string[] = [];
if (ctx.payload.web === "disabled") disallowed.push("WebFetch");
if (ctx.payload.search === "disabled") disallowed.push("WebSearch");
// both "disabled" and "restricted" block native bash
// "restricted" means use MCP bash tool instead
const bash = ctx.payload.bash;
if (bash !== "enabled") disallowed.push("Bash", "Task(Bash)");
// always block native file tools (use MCP file_read/file_write instead)
disallowed.push("Read", "Write", "Edit", "MultiEdit");
disallowed.push("Task(Read)", "Task(Write)", "Task(Edit)", "Task(MultiEdit)");
return disallowed;
}
/**
* Write MCP config file for Claude CLI.
* Returns the path to the config file.
*/
function writeMcpConfig(ctx: AgentRunContext): string {
const configDir = join(ctx.tmpdir, ".claude");
mkdirSync(configDir, { recursive: true });
const configPath = join(configDir, "mcp.json");
const mcpConfig = {
mcpServers: {
[ghPullfrogMcpName]: { type: "http", url: ctx.mcpServerUrl },
},
};
writeFileSync(configPath, JSON.stringify(mcpConfig, null, 2), "utf-8");
log.info(`» MCP config written to ${configPath}`);
return configPath;
}
async function installClaude(): Promise<string> {
const versionRange = packageJson.dependencies["@anthropic-ai/claude-agent-sdk"] || "latest";
return await installFromNpmTarball({
packageName: "@anthropic-ai/claude-agent-sdk",
version: versionRange,
executablePath: "cli.js",
});
}
export const claude = agent({
name: "claude",
install: installClaude,
run: async (ctx) => {
// install CLI at start of run
const cliPath = await installClaude();
// select model and effort level
const model = claudeEffortModels[ctx.payload.effort];
const effortLevel = claudeEffortLevels[ctx.payload.effort];
log.info(`» model: ${model}${effortLevel ? ` (effort: ${effortLevel})` : ""}`);
// build disallowedTools based on tool permissions
const disallowedTools = buildDisallowedTools(ctx);
if (disallowedTools.length > 0) {
log.info(`» disallowed tools: ${disallowedTools.join(", ")}`);
}
// write MCP config file
const mcpConfigPath = writeMcpConfig(ctx);
// build CLI args
// claude -p "prompt" --dangerously-skip-permissions --mcp-config ./mcp.json --model opus --output-format stream-json --verbose
const args: string[] = [
cliPath,
"-p",
ctx.instructions.full,
"--dangerously-skip-permissions",
"--mcp-config",
mcpConfigPath,
"--model",
model,
"--output-format",
"stream-json",
"--verbose",
];
// add --effort flag if specified (e.g. "max" for Opus 4.6)
if (effortLevel) {
args.push("--effort", effortLevel);
}
// add disallowed tools if any
if (disallowedTools.length > 0) {
args.push("--disallowedTools");
args.push(...disallowedTools);
}
log.info("» running Claude CLI...");
let stdoutBuffer = "";
let finalOutput = "";
// Track bash tool IDs to identify when bash tool results come back
const bashToolIds = new Set<string>();
const thinkingTimer = new ThinkingTimer();
const result = await spawn({
cmd: "node",
args,
cwd: process.cwd(),
env: process.env,
stdio: ["ignore", "pipe", "pipe"],
activityTimeout: 0, // disabled: process-level timeout in main.ts handles this (subprocess timeout would kill orchestrator during delegation)
onStdout: async (chunk) => {
finalOutput += chunk;
markActivity(); // reset activity timeout on any CLI output
// buffer incomplete lines across chunks (NDJSON format)
stdoutBuffer += chunk;
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 message = JSON.parse(trimmed) as SDKMessage;
markActivity(); // reset activity timeout on every event
log.debug(JSON.stringify(message, null, 2));
const handler = messageHandlers[message.type];
if (handler) {
await handler(message as never, bashToolIds, thinkingTimer);
}
} catch {
// ignore parse errors - might be non-JSON output
log.debug(`[claude] non-JSON stdout line: ${trimmed.substring(0, 200)}`);
}
}
},
onStderr: (chunk) => {
const trimmed = chunk.trim();
if (trimmed) {
log.debug(`[claude stderr] ${trimmed}`);
log.warning(trimmed);
finalOutput += trimmed + "\n";
}
},
});
if (result.exitCode !== 0) {
const errorMessage =
result.stderr ||
finalOutput ||
result.stdout ||
"Unknown error - no output from Claude CLI";
log.error(`Claude CLI exited with code ${result.exitCode}: ${errorMessage}`);
return {
success: false,
error: errorMessage,
output: finalOutput || result.stdout || "",
};
}
log.info("» Claude CLI completed successfully");
return {
success: true,
output: finalOutput || result.stdout || "",
};
},
});
type SDKMessageType = SDKMessage["type"];
type SDKMessageHandler<type extends SDKMessageType = SDKMessageType> = (
data: Extract<SDKMessage, { type: type }>,
bashToolIds: Set<string>,
thinkingTimer: ThinkingTimer
) => void | Promise<void>;
type SDKMessageHandlers = {
[type in SDKMessageType]: SDKMessageHandler<type>;
};
const messageHandlers: SDKMessageHandlers = {
assistant: (data, bashToolIds, thinkingTimer) => {
if (data.message?.content) {
for (const content of data.message.content) {
if (content.type === "text" && content.text?.trim()) {
log.box(content.text.trim(), { title: "Claude" });
} else if (content.type === "tool_use") {
// Track bash tool IDs
if (content.name === "bash" && content.id) {
bashToolIds.add(content.id);
}
thinkingTimer.markToolCall();
log.toolCall({
toolName: content.name,
input: content.input,
});
}
}
}
},
user: (data, bashToolIds, thinkingTimer) => {
if (data.message?.content) {
for (const content of data.message.content) {
if (content.type === "tool_result") {
thinkingTimer.markToolResult();
const toolUseId = (content as any).tool_use_id;
const isBashTool = toolUseId && bashToolIds.has(toolUseId);
const outputContent =
typeof content.content === "string"
? content.content
: Array.isArray(content.content)
? content.content
.map((c: any) => (typeof c === "string" ? c : c.text || JSON.stringify(c)))
.join("\n")
: String(content.content);
if (isBashTool) {
// Log bash output in a collapsed group
log.startGroup(`bash output`);
if (content.is_error) {
log.warning(outputContent);
} else {
log.info(outputContent);
}
log.endGroup();
// Clean up the tracked ID
bashToolIds.delete(toolUseId);
} else if (content.is_error) {
log.warning(`Tool error: ${outputContent}`);
} else {
// log successful non-bash tool result at debug level
log.debug(`tool output: ${outputContent}`);
}
}
}
}
},
result: async (data) => {
if (data.subtype === "success") {
const usage = data.usage;
const inputTokens = usage?.input_tokens || 0;
const cacheRead = usage?.cache_read_input_tokens || 0;
const cacheWrite = usage?.cache_creation_input_tokens || 0;
const outputTokens = usage?.output_tokens || 0;
const totalInput = inputTokens + cacheRead + cacheWrite;
log.table([
[
{ data: "Cost", header: true },
{ data: "Input", header: true },
{ data: "Cache Read", header: true },
{ data: "Cache Write", header: true },
{ data: "Output", header: true },
],
[
`$${data.total_cost_usd?.toFixed(4) || "0.0000"}`,
String(totalInput),
String(cacheRead),
String(cacheWrite),
String(outputTokens),
],
]);
} else if (data.subtype === "error_max_turns") {
log.error(`Max turns reached: ${JSON.stringify(data)}`);
} else if (data.subtype === "error_during_execution") {
log.error(`Execution error: ${JSON.stringify(data)}`);
} else {
log.error(`Failed: ${JSON.stringify(data)}`);
}
},
system: () => {},
stream_event: () => {},
tool_progress: () => {},
tool_use_summary: () => {},
auth_status: () => {},
};