mirror of
https://github.com/KeygraphHQ/shannon.git
synced 2026-07-25 22:10:54 +02:00
5ff40f8c6f
* feat(worker): migrate agent runtime from Claude Agent SDK to pi harness * feat: remove Google Vertex AI provider support * fix(worker): route Bedrock and custom-base-URL providers from env * feat(prompts): instruct agents to call submit_exploitation_queue and submit_auth_result * fix(worker): count sub-agent cost and surface compaction failures * refactor(worker): rename claude-executor to pi-executor * feat(worker): pi-event-driven output formatting * fix(worker): gate adaptive thinking to Opus models, drop CLAUDE_THINKING_LEVEL * fix(worker): restore minLength/minItems on vuln-collector schemas * feat(worker): give task sub-agent write+bash, align tool descriptions * feat(worker): add glob custom tool and route code_path globs to it * refactor(prompts): use pi tool names (task, todo_write, read, bash, glob) * refactor(prompts): drop stale MCP terminology for collector tools * refactor(prompts): drop collector server names from deliverable instructions * fix(worker): restore minLength/minItems on pre-recon and exploit collector schemas * feat(worker): load playwright-cli skill via pi resource loader * refactor(cli): remove CLAUDE_CODE_MAX_OUTPUT_TOKENS config * build: drop @anthropic-ai/claude-code from worker image * docs: remove vertex references from llms context * docs(worker): update stale sdk comments * refactor(worker): unify provider precedence between preflight and executor * feat(worker): enforce bounded bash timeouts via pi extension * ci: bump the beta release line to 2.0.0 (#356) * fix(cli): pin npx command hints to beta tag * fix: render agent deliverables before the success commit so resume preserves them (#377) * feat(cli): restructure run folder and improve terminal UX (#383) * feat: surface report at run root and nest run internals under .shannon * feat: use plain-language wording in user-facing terminal messages * feat(cli): guide users to watch scan progress and surface report path on start * docs: sync run-folder layout and CLI wording across docs and comments * feat(cli): add version command reporting package version or git SHA * feat(cli): detect TTY for interactive prompts, color, and progress output * docs: document --yes flag, version command, and tty module * fix(cli): FORCE_COLOR precedence and plain uninstall --yes output * fix(cli): respect empty NO_COLOR * fix(cli): let NO_COLOR take precedence over FORCE_COLOR * docs: mark claude-code-router integration as removed * refactor(worker): converge shared core with shannon-oss (#388) * fix(worker): port keygraph shared-core correctness fixes * refactor(worker): adopt collectors/ and ai/pi/ layout; add task budget cap and cancellation * refactor(worker): drop inconsistent Collector "Server" suffix * refactor(worker): drop unused providerConfig/apiKey seams, resolve credentials from env only * refactor(worker): port oss code_path pattern expansion + external_directory allow * fix(worker): preserve dotfile paths in code_path avoid patterns (.env no longer stripped to env) * feat(worker): render Unprocessed Vulnerabilities section in exploit deliverable (align with oss) * feat(worker): request set_blind_spots for all vuln classes (align auth/ssrf with production prompts) * refactor(worker): adopt unified permissionSystem* naming and helper layout * refactor(worker): inline blind_spots into vuln deliverable section array * chore(worker): drop unused zod dependency (tree is typebox-native) * fix(worker): normalize base32 TOTP secret to accept padding and whitespace * refactor(worker): adopt shared toolResult helper and flatSchema naming in collectors * refactor(worker): use undefined over null in queue-schema builders * docs(worker): converge renderer/collector doc comments to current pi terminology * refactor(worker): adopt schema.ts cleanInput/stringEnum helpers in collectors * feat(worker): converge exploit-collector/renderer with vendored; capture and render overview for blocked findings * refactor(worker): converge session-tools/pipeline/exploitation-checker with vendored * refactor(worker): converge task-tool usage reporting with vendored onUsage callback * refactor(worker): converge structured output onto a submitTool executor channel * docs(worker): expand exploit-renderer docstring to match shannon-oss * docs(worker): adopt richer vuln-renderer docstring from shannon-oss * docs(worker): neutralize billing-detection wording for shannon-oss parity * fix(worker): verify checkpoint hash in the deliverables clone being reset * fix(worker): fail fast on malformed exploitation queue JSON * fix(worker): honor retryable flag when classifying exploitation-queue check failures * fix(worker): fail fast on corrupted session.json in run-scope validation * feat(worker): propagate Temporal cancellation signal into agent and auth pi sessions * fix(worker): mark exploit agent complete when exploitation is skipped so resume skips it * prompts: drop scan description from executive report prompt * refactor(worker): add createGenericSubmitTool for raw JSON-schema submit tools * refactor(worker): gate playwright-cli skill to browser agents via skillsOverride (adopt shannon-oss mechanism) * docs(worker): correct formatLogTime comment to UTC to match toISOString * refactor(worker): converge queue-schemas with shannon-oss (guarded count, decl order) * refactor(worker): converge task-tool with shannon-oss (byte-identical; modelRegistry optional) * fix(worker): use replaceLiteral for all prompt value insertions to prevent $-mangling * fix(worker): classify agent execution failures by error type instead of hardcoding validation * fix(worker): cap auth-failure detail at 250 chars to match shannon-oss * style(worker): apply biome formatting * refactor(worker): remove per-session task delegation cap from task tool * style(cli): collapse usage hint now that the beta tag is gone * chore: mark the pi harness migration as a breaking change BREAKING CHANGE: Google Vertex AI is no longer a supported provider. The CLAUDE_CODE_USE_VERTEX, ANTHROPIC_VERTEX_PROJECT, CLOUD_ML_REGION, and GOOGLE_APPLICATION_CREDENTIALS environment variables, along with the use_vertex, vertex_project, and cloud_ml_region config.toml keys, are removed. Vertex users must switch to Anthropic, AWS Bedrock, or a custom Anthropic-compatible base URL. The CLAUDE_CODE_MAX_OUTPUT_TOKENS environment variable and the max_output_tokens config.toml key are also removed.
61 lines
2.4 KiB
TypeScript
61 lines
2.4 KiB
TypeScript
// Copyright (C) 2025 Keygraph, Inc.
|
|
//
|
|
// This program is free software: you can redistribute it and/or modify
|
|
// it under the terms of the GNU Affero General Public License version 3
|
|
// as published by the Free Software Foundation.
|
|
|
|
import { defineTool, type ToolDefinition } from '@earendil-works/pi-coding-agent';
|
|
import { Type } from 'typebox';
|
|
|
|
/**
|
|
* A pi custom submit tool plus the captured payload it records.
|
|
*
|
|
* pi ships no JSON-schema output format, so an agent that must return structured
|
|
* data does so by calling a purpose-built TypeBox tool. This bundles that tool
|
|
* with its capture accessor and the directive that instructs the model to call
|
|
* it. The executor owns the wiring — it registers the tool, appends the
|
|
* directive to the prompt, and reads `getCaptured()` back as `structuredOutput`
|
|
* — so callers never assemble it by hand.
|
|
*/
|
|
export interface CapturedSubmitTool {
|
|
readonly tool: ToolDefinition;
|
|
readonly getCaptured: () => unknown | undefined;
|
|
readonly directive?: string;
|
|
}
|
|
|
|
/**
|
|
* Build a `submit_result` tool from a raw JSON Schema, for agents whose result
|
|
* shape is not one of the built-in per-agent schemas (e.g. an out-of-tree agent
|
|
* with its own verdict schema). pi validates the tool call against `schema`
|
|
* before `execute()` runs, so a captured payload is already schema-valid — no
|
|
* separate validation pass is needed.
|
|
*/
|
|
export function createGenericSubmitTool(schema: Record<string, unknown>): CapturedSubmitTool {
|
|
let captured: unknown | undefined;
|
|
return {
|
|
tool: defineTool({
|
|
name: 'submit_result',
|
|
label: 'Submit Result',
|
|
description: 'Return your final structured answer. Call exactly once as your last action.',
|
|
promptSnippet: 'submit_result: deliver your structured answer (call once)',
|
|
promptGuidelines: [
|
|
'You MUST call submit_result exactly once as your final action.',
|
|
'Fill every required parameter. Do not output JSON as text.',
|
|
],
|
|
parameters: Type.Unsafe(schema),
|
|
async execute(_toolCallId, params) {
|
|
captured = params;
|
|
return {
|
|
content: [{ type: 'text' as const, text: 'Result submitted.' }],
|
|
details: params,
|
|
terminate: true,
|
|
};
|
|
},
|
|
}),
|
|
getCaptured: () => captured,
|
|
directive:
|
|
'\n\nYou MUST call the submit_result tool exactly once as your final action ' +
|
|
'to deliver your structured answer. Do not output JSON as text. Fill every required parameter.',
|
|
};
|
|
}
|