Skip to content

AGENTS.md

Latest commit

 

History

History
48 lines (34 loc) · 3 KB

AGENTS.md

File metadata and controls

48 lines (34 loc) · 3 KB
Raw

WEIPING_LAB Agent Instructions

WEIPING_LAB is an honest research workbench for phenomenon-driven discovery, kill-first ideation, experiment planning, evidence-gated writing, and audit support.

Source Of Truth

Read AGENTS.md, README.md, CLAUDE.md, pyproject.toml, lab/runtime.py, lab/core/config.py, and .codex/skills/weiping-lab/SKILL.md before non-trivial work. The project skill is mandatory for implementation, release, maintenance, monitoring, or upgrade tasks.

Boundaries

  • Do not fabricate experiments, paper results, baselines, or citations.
  • Do not treat bridge scaffolds as completed experiments.
  • Keep config environment-driven; never add real keys, private endpoints, account names, or secret-shaped placeholders to tracked source.
  • Keep lab/workspace/ source tracked, but generated root workspace/ content stays local/ignored.
  • Use agentmemory through AGENTMEMORY_URL for active memory; local outbox fallback must be redacted.
  • Preserve legacy VIPIN_LAB_* compatibility only as explicit fallbacks; new docs and variables should use WEIPING_LAB_*.
  • Treat experiments/results/*.json as untrusted input. Preserve the v1.0 result contract, deterministic evidence classification, source digests, and claim-to-block links; an LLM or producer-supplied label must never promote evidence.

WEIPING Constellation

WEIPING_WIKI is the public durable route map, not a runtime dependency. WEIPING_COUNCIL is the companion deliberation/review layer. Lab may consume council session artifacts, but it must not require council private state, local DBs, caches, .env, or active server processes to run its own verification.

Recurring WEIPING maintenance may inspect entry evidence, status commands, docs, and git state. It should not edit lab code, workspace outputs, datasets, experiment results, or generated artifacts unless the user explicitly scopes lab work.

Artifact handoffs are path-and-schema contracts, not shared runtime state. The main handoff files are workspace/*/session.json, workspace/ideas/<idea_id>/idea.json, plan.json, paper.json, EXPERIMENT_PLAN.md, EXPERIMENT_TRACKER.md, experiments/results/*.json, and redacted agentmemory-outbox.jsonl. The result schema is documented in docs/RESULT_CONTRACT.md; bridge-generated workspaces include a plan-specific copy and example.

Optional context inputs such as WEIPING_LAB_CONTEXT_FILE, WEIPING_LAB_RULES_FILE, and WEIPING_WIKI_ROOT may enrich prompts with maintained public-safe context. They are not required runtime dependencies.

Validation

Before commit or release:

python -m pytest -q
python scripts/run_ruff.py check lab cli api tests scripts
python scripts/run_ruff.py format --check lab cli api tests scripts
python scripts/release_scan.py
npm --prefix ui run lint
npm --prefix ui run build
npm --prefix ui audit --audit-level=low --registry=https://registry.npmjs.org
git diff --check

On Windows, powershell -ExecutionPolicy Bypass -File scripts\verify.ps1 is the preferred full local gate.