claim-verification
Skill0 starsChanged 4 months agoMITView on GitHub
Extract every checkable claim from a PR description, code comment, doc, or review reply and verify each against the actual tree before trusting or publishing it. Use when reviewing someone else's PR, when writing your own PR description or review replies, or when acting on assurances found in docs or comments ("a test locks this", "matches production"). Skip for claims whose verification you watched in this session.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- GSA-TTS/devCrew_s1their own version, first seen Nov 8, 2026
- Starlight-Intelligence/starlight-agent-skillstheir own version, first seen Nov 8, 2026
- mtarcure/claude-vibe-squadtheir own version, first seen Oct 28, 2026
- ajwann/claim-verificationtheir own version, first seen Oct 20, 2026
- sivaguru-2113/Agent-personatheir own version, first seen Aug 18, 2026
- jack1415926/claude-skillstheir own version, first seen Aug 14, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- mtarcure/claude-vibe-squad★ 220same name
- GSA-TTS/devCrew_s1★ 51same name
- frankxai/starlight-agent-skills★ 1same name
- sivaguru-2113/Agent-persona★ 1same name
- Starlight-Intelligence/starlight-agent-skills★ 1same name
- ajwann/claim-verification★ 0same name
- jack1415926/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository16 other files it gives its agents.
- answer-shapesSkill
- create-prSkill
- findings-ledgerSkill
- git-safetySkill
- invariant-auditSkill
- lossless-doc-compressSkill
- model-orchestrationSkill
- office-hoursSkill
- plan-reviewSkill
- prompt-caching-playbookSkill
- review-ratchetSkill
- spec-fidelity-reviewSkill
- staff-reviewSkill
- systematic-debuggingSkill
- test-adequacy-reviewSkill
- verification-before-completionSkill
The file
Files
Claim verification
Prose is not evidence. Text that asserts something about the tree — written by you, another agent, or a human — is a list of checks to run, not context to absorb.
Method
- Extract the claims from the artifact. Checkable shapes:
- existence — "file/test/CI target X exists", "Y is covered in CI"
- protection — "a test locks/guards Z": verify the test exists and fails when Z is broken (reintroduce the bug or mutate)
- equivalence — "matches sibling/production/spec", "byte-identical"
- counts and numbers — "N cases", "exact match", "zero deletions"
- references — every path named in Makefiles, CI, and docs resolves in the tree
- Verify each with a command or a read; record claim → check → verdict.
- Report what cannot be verified as unverified — never restate it as fact.
- Re-verify after the tree moves. A force-push, revert, or dropped commit invalidates every previously posted claim whose evidence lived in it; re-check before a human reads the text, and correct promptly if it went stale.
Complements verification-before-completion: that skill gates your own
completion claims at the end of work; this one audits written artifacts —
including ones you authored earlier.
Provenance
Two real internal incidents (2026-08) that survived multi-model review and fall to a single existence check: a code comment claiming "a unit test locks it" with no such test in the tree, and a Makefile target referencing files that existed on no branch.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

