verify-claimlisted

# /verify-claim — Research Rigor Enforcement **Purpose**: Stop unverified factual claims (counts, names, versions, dates, quotes) from reaching the reader. Verify or mark `[unverified]` — never guess, never interpolate. ## When to Run - Before publishing any research memo, strategy doc, or competitive analysis - After drafting a section that contains numbers, names, or versions - When asked to "fact-check" or "verify" any output - Proactively during long-form analysis sessions, every ~500 words ## Inputs - A draft (file path, pasted text, or current conversation output) - Optional: list of primary sources to check against (repo paths, URLs, docs) ## Process ### 1. Extract Claims Parse the draft and produce a numbered list of every **factual assertion**. A factual assertion is anything where wrong = wrong, not opinion. Includes: - Counts (e.g., "9 agents", "3 repos", "187 commits") - Named entities (people, products, repos, commands, files, libraries) - Versions (model IDs, package versions, dates) - Direct or indirect quotes - Causal claims about specific events ("this broke because X") Skip: opinions, recommendations, hypotheticals. ### 2. Verify Each Claim For each claim, attempt verification in this order: 1. **Repo evidence** — `grep`, file `Read`, `git log` for claims about code/commits/configs 2. **Live source** — `WebFetch` for claims about external URLs/APIs/docs 3. **Authoritative reference** — official docs (use `context7` MCP for library/SDK claims) 4.