Machine-verifiable claims for design docs and agent-written proposals. Citations your CI can re-check.
Agent Claude Code
Use when you need a focused, parallel-dispatchable architectural review of a diff, a planned change, or an OpenSpec proposal against this repo's cross-cutting invariants — the doctrine written as prose in CLAUDE.md, spec/.md, and openspec/project.md, which has no .claude/rules/.md file of its own. Returns a structured…
Agent Claude Code
Use when a diff, planned change, or OpenSpec proposal touches the checker kernel — packages/claims/src/checkClaims.ts, witness.ts, wiring.ts, rules.ts, or config.ts — and you need a review of whether the change respects the kernel's own semantics: which verdict union a new verdict belongs to, whether its pass/fail…
Agent Claude Code
Use as the FINAL stage of a proposal-to-pr run to record what happened, as one retrospective file under .claude/retrospectives/. Dispatched fresh, having NOT done the work, so it reads artefacts — the pipeline state file, review-evidence.md and its ## Probe — stage 2 score, progress.md, git history — rather than the…
Agent Claude Code
Use when you need a focused, parallel-dispatchable rule-compliance review of a diff, a planned change, or an OpenSpec proposal against the project rule files in .claude/rules/. Returns a structured [blocker] / [concern] / [looks-good] report. Especially useful inside the proposal-to-pr pipeline where rule-audit must…
Agent Claude Code
Use when a diff, planned change, or OpenSpec proposal touches packages/claims/src//.ts, packages/kit/src//.ts, spec/fixtures//.jsonl, or .github/workflows/.yml, and you need a review of whether the change's test and fixture coverage actually proves what it claims — not whether the kernel semantics themselves are…
Agent Claude Code
An agent that exists, so the skill dispatching to it resolves.