From pulseengine-claude
This skill should be used whenever findings, audits, code-review results, claims, or analysis output need to be validated before reporting — including "verify this", "double-check this", "audit", "is this actually true", "before I report this", "before we merge this", or whenever an agent's summary needs independent confirmation. ALWAYS use this skill before delivering non-trivial inspection results, before claiming a property holds, and whenever agent-produced hashes, digests, versions, file paths, or flag names appear in a report.
How this skill is triggered — by the user, by Claude, or both
Slash command
/pulseengine-claude:clean-room-verificationThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Any time you're about to report findings from an investigation, audit, code review, or analysis — *especially* when those findings include claims about behavior, state, version pins, hashes, file paths, flag names, or "everything is green / done." Also fires when reviewing another agent's deliverable: a summary describes intent, not what landed.
Any time you're about to report findings from an investigation, audit, code review, or analysis — especially when those findings include claims about behavior, state, version pins, hashes, file paths, flag names, or "everything is green / done." Also fires when reviewing another agent's deliverable: a summary describes intent, not what landed.
This is the smithy ritual in PulseEngine vocabulary. The point is to catch hallucinations and over-claims before they ship.
Write the findings as discrete falsifiable claims. Each one should be specific enough that an independent checker can confirm, refute, or say "cannot verify." Examples of good claims:
foo::bar in crates/foo/src/lib.rs returns Result<(), Error> and propagates errors via ?."docker.io/nixos/nix@sha256:abcdef... is pullable from a fresh client."rivet check on this branch reports zero unsatisfied predicates."wohl-ota pass with cargo kani."Avoid vague claims like "the refactor is clean" or "tests pass" — those aren't checkable.
Spawn a clean-room verification subagent. Brief it cold — no inherited framing, no access to the original reasoning, no narrative about why the work was done. Give it only:
confirm, refute, or cannot-verify per claim. Do not guess. cannot-verify is a valid and preferred answer over a guess."Treat agent-produced artifacts as unverified claims. Hashes, digests, version pins, file paths, flag names, symbol names — even when produced by a tool that should be authoritative — count as claims until checked against the real artifact. The verifier should pull the image, grep the binary, read the file, run the command.
Reconcile. Compare the verifier's confirm/refute/cannot-verify against your draft findings:
refute → the claim is wrong; rewrite or drop it.cannot-verify → either add the evidence the verifier needs, or downgrade the claim to "this is suspected, not verified."confirm → the claim ships as-is.Report. Include the verifier's verdict alongside the claim, so the reader can see what was independently checked vs. what was asserted.
oracle-gate-a-change] — mechanical oracle preferred.oracle-gate-a-change and pulseengine-feature-loop both point here for their verify step. The pattern is single-source-of-truth here; if you're inlining the procedure elsewhere, you're duplicating a known dependency.
Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
npx claudepluginhub pulseengine/pulseengine.eu --plugin pulseengine-claude