Introduces three new weekly agents that form a triage→plan→build pipeline:
- proposal-triage-agent: evaluates proposals, dead ends, and coverage gaps;
produces prioritized backlog in src/backlog/BACKLOG.md
- plan-agent: creates concrete implementation plans from top backlog items
in src/plans/PLAN-*.md
- builder-agent: executes approved plans, makes code changes, verifies
acceptance criteria
Also includes:
- Architecture proposal document in docs/proposals/
- Seeded initial backlog with 11 items from diagnosed dead ends
- Updated roster.json and weekly-scheduler.md (24 weekly agents)
- All three validators pass (prompt-contract, scheduler-roster, flow-parity)
https://claude.ai/code/session_01FLP7ibXd5QPG517g1kYCHL
- Moved `withTimeout` implementation to `src/lock-utils.mjs` and exported it correctly.
- Removed duplicate local definitions of `withTimeout`, `mergeRelayList`, and `secureRandom` in `src/lock-ops.mjs` and `src/lock-publisher.mjs`.
- Fixed imports in `src/lock-ops.mjs`, `src/lock-publisher.mjs`, and `src/relay-health.mjs` to use `src/lock-utils.mjs` consistently.
- Updated `test/utils.test.mjs` and `test/utils-async.test.mjs` to fix broken imports and test definitions.
- Verified `npm run lock:complete` and `npm run scheduler:daily` execute correctly.
- Updated `src/todo/TODO_2026-02-20T03-00-00Z.md` to mark verification task as completed.
- Verified `src/relay-health.mjs` does not use `MINUTE_MS` locally, despite mysterious CI error. Confirmed imports are correct.
Co-authored-by: PR0M3TH3AN <[email protected]>
The `perf-agent` prompt contained detailed instructions to audit `/content` upload documentation and features. Since no such features or documentation exist in the repository (as confirmed by `content-audit-agent`), these instructions were causing confusion and wasted effort.
This change removes:
- The "Docs audit" sub-workflow and its associated steps.
- References to `/content` and upload/contribution docs in the Mission, Focus, and Goals.
- The "docs: `/content` upload alignment" PR template.
This aligns the agent's instructions with the actual codebase capabilities.
Co-authored-by: PR0M3TH3AN <[email protected]>
- run-selected-prompt.mjs: set SCHEDULER_MEMORY_FILE env var before spawning each agent so
store.mjs has a well-known path to read from (memory-updates/<ts>__<agent>.md)
- retrieve.mjs: format memories.md using m.summary + date/tags in readable markdown sections
instead of raw m.content blobs; agents can now scan injected context at a glance
- store.mjs: remove semantically-empty fallback placeholder events that polluted the store
and degraded retrieval quality; also fix default cadence 'weekly' -> 'daily' to match retrieve
- scheduler-flow.md: replace vague "write insights" instruction with a concrete three-section
template (Key findings / Patterns / Warnings) and clarify that $SCHEDULER_MEMORY_FILE is
set by the scheduler automatically
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
- Added `.github/workflows/ci.yml` to run build, lint, and test on PRs/push to main.
- Updated `.github/workflows/ci-resilience-nightly.yml` to use Node.js 22.
- Fixed lint errors in `src/services/governance/index.js` (no-empty, preserve-caught-error).
- Added `governance-agent` to daily roster and scheduler prompt to fix `npm test` scheduler validation failure.
Co-authored-by: PR0M3TH3AN <[email protected]>
Implement a comprehensive governance system to manage agent prompt changes safely.
Key features:
- **Proposal Workflow:** Prompt changes are now "proposals" stored in `src/proposals/` with metadata and diffs.
- **Governance Agent:** A new `governance-agent` reviews and applies proposals using `scripts/governance/process-proposals.mjs`.
- **Validation:** Strict validation ensures changes are limited to allowed directories (`src/prompts/daily|weekly`) and preserve required headers.
- **Lock Enhancements:** Lock events now include `promptHash`, `promptPath`, and `gitCommit` for full auditability.
- **Rollback:** New `torch-lock rollback` command restores prompts from local history (`.torch/prompt-history/`) or git.
- **CLI:** Added `proposal` (create, list, show, apply, reject) and `rollback` commands.
This establishes a "Stage -> Evaluate -> Apply" gate for prompt evolution, preventing silent drift and ensuring constitutional stability.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Refactored `test/memory-scheduler.test.mjs` to use `mock.timers` for deterministic execution, replacing flaky `wait` logic.
- Strengthened `test/lock-ops.test.mjs` with edge case tests for `parseLockEvent` handling invalid JSON and missing fields.
- Fixed a security vulnerability in `src/dashboard.mjs` where sensitive configuration (e.g., rosters, auth) was exposed in `/torch-config.json`.
- Updated `src/prompts/daily/test-audit-agent.md` to comply with the canonical prompt contract.
- Added `reports/test-audit/test-integrity-note-2025-02-14.yaml` detailing test changes and anti-cheat measures.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Added "Clean Repo Standard" to `AGENTS.md` and `CLAUDE.md`, explicitly forbidding root-level logs/artifacts.
- Moved clutter files (`test_output_full.txt`, `benchmark-after.txt`, `dashboard.log`, etc.) to `test_logs/`, `task-logs/`, and `reports/performance/`.
- Moved `INITIAL_BASELINE.md` and other baseline files to `reports/performance/`.
- Updated `src/prompts/daily/perf-agent.md` and `src/prompts/daily/const-refactor-agent.md` to reference the new location of `INITIAL_BASELINE.md`.
- Updated `.gitignore` to ignore logs and temporary artifacts in the new directories.
- Moved `DIAGNOSIS.md` to `reports/performance/`.
- Moved `perf/` data files to `reports/performance/`.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Created `reports/` directory with subdirectories for each category (audit, performance, security, etc.).
- Moved existing reports to the new `reports/` structure.
- Updated `AGENTS.md` to define the "Report Location Standard".
- Updated all agent prompts in `src/prompts/` to:
- Output reports to the new locations.
- Fix incorrect script paths (replaced `torch/scripts/` with `scripts/`).
- Updated `load-test-agent.md` to produce a Markdown summary report.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Created `AGENTS.md` and `CLAUDE.md` to define test integrity rules and behavior-based validation.
- Updated `TORCH.md` to include "Scenario-First Tests & Test Integrity" and the "Test Integrity & Scenario Spec Agent" role.
- Updated `src/prompts/daily/test-audit-agent.md` to adopt the new "Enforcer" role, prioritizing test fidelity over CI green state.
- Preserved existing scheduler contract headers in the agent prompt.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Updated `src/prompts/daily/test-audit-agent.md` to instruct the agent to run only unit tests during flakiness detection.
- Explicitly excluded composite scripts (e.g. `npm test` with lint/build) and slow integration/E2E tests from the repeated test loop.
- This prevents timeouts caused by running the full test suite (including network-dependent tests) multiple times.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Modified `scripts/agent/run-scheduler-cycle.mjs` to generate a unique `memory-updates/<timestamp>__<agent>.md` path and pass it via `SCHEDULER_MEMORY_FILE` env var.
- Updated `scripts/memory/store.mjs` to prioritize `SCHEDULER_MEMORY_FILE` for ingestion.
- Updated `src/prompts/scheduler-flow.md` to instruct agents to write to the unique memory file.
- Updated `src/prompts/daily/torch-garbage-collection-agent.md` to include `memory-updates/*.md` in the cleanup scope.
- Moved existing `memory-update.md` to `memory-updates/legacy-memory-update-<timestamp>.md` to clear the root.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Merged AGENTS.md, CLAUDE.md, and src/docs/TORCH.md into a single TORCH.md in the root.
- Updated src/ops.mjs to copy TORCH.md during init/update and correctly transform path references.
- Updated src/prompts/META_PROMPTS.md and src/prompts/scheduler-flow.md to reference TORCH.md.
- Deleted obsolete AGENTS.md, CLAUDE.md, and src/docs/TORCH.md.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Extracted `parseArgs` function from `src/lib.mjs` to `src/cli-parser.mjs`.
- Added unit tests for argument parsing logic in `test/cli-parser.test.mjs`.
- Updated `src/lib.mjs` to import `parseArgs`.
- Fixed roster drift in `src/prompts/weekly-scheduler.md` to match `roster.json`.
- Verified with `npm test` and `npm run lint`.
- Completed weekly scheduler task for `refactor-agent`.
Co-authored-by: PR0M3TH3AN <[email protected]>
- Update weekly-scheduler.md to include prompt-fixer-agent
- Generate run artifacts for scheduler-update-agent
Co-authored-by: PR0M3TH3AN <[email protected]>
This commit introduces a new agent, `prompt-fixer-agent`, designed to automatically remediate safety findings identified by the `prompt-safety-agent`.
- Added `src/prompts/weekly/prompt-fixer-agent.md`: Defines the agent's mission to parse `artifacts/prompt-safety-audit.md` and inject missing `FAILURE MODES` and `EXIT CRITERIA` sections into flagged agent prompts.
- Updated `src/prompts/roster.json`: Added `prompt-fixer-agent` to the `weekly` roster, scheduled to run after `prompt-safety-agent`.
This agent ensures that all prompts in the repository maintain a high standard of safety and resilience by providing automated fixes for common omissions.
Co-authored-by: PR0M3TH3AN <[email protected]>
