Cursor rule
A clean Failed is a first-class research result — it prevents every future session from re-walking the same dead end.
67 1mo ago A 389 tokens
Cursor rule
Honest measurement is the foundation: the registry is only ground truth if the gates behind it cannot be gamed.
67 1mo ago A 442 tokens
Cursor rule
One canonical implementation per concern. Extend canonical systems via their declared interface — never build a parallel copy.
67 1mo ago A 239 tokens
Cursor rule
Understand before code. Agent entry point: AGENTS.md at the repo root.
67 1mo ago A 296 tokens
Cursor rule
Project skills (copy skills/ into .cursor/skills/).
67 1mo ago A 420 tokens
Cursor rule
No green verification → feature is not implemented. No measured positive impact → feature stays off (or is deleted).
67 1mo ago A 462 tokens
Cursor rule
Cursor rule "verification-memory" from agentic-commerce-lab/pawl, covering verification memory, mandatory and boundaries.
67 1mo ago A 384 tokens
Skill Claude CodeCodex
Implements verification claims — register.py, initfolder.py, frozen gates, tools/run.py, honest promotion. Use when adding a feature, creating a claim, promoting Planned to Verified/Failed, or following the lifecycle before writing feature code.
67 1mo ago A 52 tokens
Skill Claude CodeCodex
Fast claim-harness iteration — reuse flags, single-job runs, cache discipline, honest labeling of dev shortcuts. Use when developing or debugging a claim gate before the official tools/run.py pass.
67 1mo ago A 42 tokens
Skill Claude CodeCodex
Documents honest-negative verification outcomes when a claim fails — names the exact missing capability, updates the registry to Failed, sweeps dependent Planned claims, no shadow fixes. Use when tools/run.py exits 1, a claim fails its gates, or an experiment is terminal-negative although the mechanism is sound.
67 1mo ago A 63 tokens
Skill Claude CodeCodex
Coordinates parallel agent sessions and git worktrees for verification work without colliding on registry.csv or shared config. Use when launching multiple background workers, parallel verification waves, worktree isolation, or merging claim branches.
67 1mo ago A 44 tokens
Skill Claude CodeCodex
Fact-checks plans and architecture claims against registry.csv, claim lastrun.json evidence, and real code. Use when analyzing plans, verifying strategy documents, judging if a claim exists, checking Planned vs Verified status, or answering "does the code actually match this plan?".
67 1mo ago A 58 tokens
Skill Claude CodeCodex
Reading map for substantive work — which big-picture and status docs to load before coding, verifying, or analyzing plans. Use when starting substantive edits, exploring architecture, onboarding to the repo, or before touching core modules.
67 1mo ago A 47 tokens
Skill Claude CodeCodex
Queries the generated verification memory (tools/memory.py) — claims (ground truth) + plans (proposals), research fronts, per-claim dossiers, plan-to-claim links, BM25 search. Use when asking what was already tried, what plans exist for a topic, claim neighborhood, or before registering a new claim.
67 1mo ago A 69 tokens