Files
shannon/apps/worker/src/session-manager.ts
T
ezl-keygraph 5ff40f8c6f feat(worker): migrate agent runtime from Claude Agent SDK to pi harness (#389)
* 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.
2026-07-16 19:13:13 +05:30

226 lines
8.0 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 { fs, path } from 'zx';
import type { ActivityLogger } from './types/activity-logger.js';
import type { AgentDefinition, AgentName, AgentValidator, PlaywrightSession, VulnType } from './types/index.js';
// Agent definitions according to PRD
export const AGENTS: Readonly<Record<AgentName, AgentDefinition>> = Object.freeze({
'pre-recon': {
name: 'pre-recon',
displayName: 'Pre-recon agent',
prerequisites: [],
promptTemplate: 'pre-recon-code',
deliverableFilename: 'pre_recon_deliverable.md',
modelTier: 'large',
},
recon: {
name: 'recon',
displayName: 'Recon agent',
prerequisites: ['pre-recon'],
promptTemplate: 'recon',
deliverableFilename: 'recon_deliverable.md',
},
'injection-vuln': {
name: 'injection-vuln',
displayName: 'Injection vuln agent',
prerequisites: ['recon'],
promptTemplate: 'vuln-injection',
deliverableFilename: 'injection_analysis_deliverable.md',
},
'xss-vuln': {
name: 'xss-vuln',
displayName: 'XSS vuln agent',
prerequisites: ['recon'],
promptTemplate: 'vuln-xss',
deliverableFilename: 'xss_analysis_deliverable.md',
},
'auth-vuln': {
name: 'auth-vuln',
displayName: 'Auth vuln agent',
prerequisites: ['recon'],
promptTemplate: 'vuln-auth',
deliverableFilename: 'auth_analysis_deliverable.md',
},
'ssrf-vuln': {
name: 'ssrf-vuln',
displayName: 'SSRF vuln agent',
prerequisites: ['recon'],
promptTemplate: 'vuln-ssrf',
deliverableFilename: 'ssrf_analysis_deliverable.md',
},
'authz-vuln': {
name: 'authz-vuln',
displayName: 'Authz vuln agent',
prerequisites: ['recon'],
promptTemplate: 'vuln-authz',
deliverableFilename: 'authz_analysis_deliverable.md',
},
'injection-exploit': {
name: 'injection-exploit',
displayName: 'Injection exploit agent',
prerequisites: ['injection-vuln'],
promptTemplate: 'exploit-injection',
deliverableFilename: 'injection_exploitation_evidence.md',
},
'xss-exploit': {
name: 'xss-exploit',
displayName: 'XSS exploit agent',
prerequisites: ['xss-vuln'],
promptTemplate: 'exploit-xss',
deliverableFilename: 'xss_exploitation_evidence.md',
},
'auth-exploit': {
name: 'auth-exploit',
displayName: 'Auth exploit agent',
prerequisites: ['auth-vuln'],
promptTemplate: 'exploit-auth',
deliverableFilename: 'auth_exploitation_evidence.md',
},
'ssrf-exploit': {
name: 'ssrf-exploit',
displayName: 'SSRF exploit agent',
prerequisites: ['ssrf-vuln'],
promptTemplate: 'exploit-ssrf',
deliverableFilename: 'ssrf_exploitation_evidence.md',
},
'authz-exploit': {
name: 'authz-exploit',
displayName: 'Authz exploit agent',
prerequisites: ['authz-vuln'],
promptTemplate: 'exploit-authz',
deliverableFilename: 'authz_exploitation_evidence.md',
},
report: {
name: 'report',
displayName: 'Report agent',
prerequisites: ['injection-exploit', 'xss-exploit', 'auth-exploit', 'ssrf-exploit', 'authz-exploit'],
promptTemplate: 'report-executive',
deliverableFilename: 'comprehensive_security_assessment_report.md',
},
});
// Phase names for metrics aggregation
export type PhaseName = 'pre-recon' | 'recon' | 'vulnerability-analysis' | 'exploitation' | 'reporting';
// Map agents to their corresponding phases (single source of truth)
export const AGENT_PHASE_MAP: Readonly<Record<AgentName, PhaseName>> = Object.freeze({
'pre-recon': 'pre-recon',
recon: 'recon',
'injection-vuln': 'vulnerability-analysis',
'xss-vuln': 'vulnerability-analysis',
'auth-vuln': 'vulnerability-analysis',
'authz-vuln': 'vulnerability-analysis',
'ssrf-vuln': 'vulnerability-analysis',
'injection-exploit': 'exploitation',
'xss-exploit': 'exploitation',
'auth-exploit': 'exploitation',
'authz-exploit': 'exploitation',
'ssrf-exploit': 'exploitation',
report: 'reporting',
});
// Factory function for vulnerability queue validators.
//
// The analysis_deliverable.md is rendered via the writeDeliverable hook, which
// AgentExecutionService runs after validateAgentOutput but before the success
// commit — so a "both files exist" check here would race the renderer. The
// validator only checks queue.json, written by the submit-tool path in
// agent-execution.ts before this validator runs.
function createVulnValidator(vulnType: VulnType): AgentValidator {
return async (sourceDir: string, logger: ActivityLogger): Promise<boolean> => {
const queueFile = path.join(sourceDir, `${vulnType}_exploitation_queue.json`);
const queueExists = await fs.pathExists(queueFile);
if (!queueExists) {
logger.warn(`Queue validation failed for ${vulnType}: ${vulnType}_exploitation_queue.json missing`);
return false;
}
return true;
};
}
// Exploitation agents — the evidence deliverable is rendered via the writeDeliverable
// hook after the agent succeeds (before the success commit), so a file-existence check
// here would race the renderer.
//
// VulnType is kept in the import surface for createVulnValidator above; this factory
// returns a no-op validator parameterized only for symmetry with the vuln-side factory.
function createExploitValidator(_vulnType: VulnType): AgentValidator {
return async (): Promise<boolean> => true;
}
// Playwright session mapping - assigns each agent to a specific session for browser isolation
// Keys are promptTemplate values from AGENTS registry
export const PLAYWRIGHT_SESSION_MAPPING: Record<string, PlaywrightSession> = Object.freeze({
// Runs before any agent — non-concurrent, so agent1 is safe to share
'validate-authentication': 'agent1',
// Phase 1: Pre-reconnaissance
'pre-recon-code': 'agent1',
// Phase 2: Reconnaissance
recon: 'agent2',
// Phase 3: Vulnerability Analysis (5 parallel agents)
'vuln-injection': 'agent1',
'vuln-xss': 'agent2',
'vuln-auth': 'agent3',
'vuln-ssrf': 'agent4',
'vuln-authz': 'agent5',
// Phase 4: Exploitation (5 parallel agents - same as vuln counterparts)
'exploit-injection': 'agent1',
'exploit-xss': 'agent2',
'exploit-auth': 'agent3',
'exploit-ssrf': 'agent4',
'exploit-authz': 'agent5',
// Phase 5: Reporting
'report-executive': 'agent3',
});
// Direct agent-to-validator mapping - much simpler than pattern matching
export const AGENT_VALIDATORS: Record<AgentName, AgentValidator> = Object.freeze({
// Pre-reconnaissance agent — skipped tools surface as renderer placeholders, not
// activity failures. The deliverable file is written by the renderer after the agent
// succeeds, so a file-existence check here would race the renderer.
'pre-recon': async (): Promise<boolean> => true,
// Reconnaissance agent — validation lives in runReconAgent post-processing.
// The deliverable file is written by the renderer after the agent succeeds, so a
// file-existence check here would race the renderer.
recon: async (): Promise<boolean> => true,
// Vulnerability analysis agents
'injection-vuln': createVulnValidator('injection'),
'xss-vuln': createVulnValidator('xss'),
'auth-vuln': createVulnValidator('auth'),
'ssrf-vuln': createVulnValidator('ssrf'),
'authz-vuln': createVulnValidator('authz'),
// Exploitation agents
'injection-exploit': createExploitValidator('injection'),
'xss-exploit': createExploitValidator('xss'),
'auth-exploit': createExploitValidator('auth'),
'ssrf-exploit': createExploitValidator('ssrf'),
'authz-exploit': createExploitValidator('authz'),
// Executive report agent
report: async (sourceDir: string, logger: ActivityLogger): Promise<boolean> => {
const reportFile = path.join(sourceDir, 'comprehensive_security_assessment_report.md');
const reportExists = await fs.pathExists(reportFile);
if (!reportExists) {
logger.error('Missing required deliverable: comprehensive_security_assessment_report.md');
}
return reportExists;
},
});