← ClaudeAtlas

app-claim-checklisted

Guards every substantive claim in an <app> deliverable: each claim cites a resolvable evidence id, carries a confidence tier, and open questions stay in a ledger until cited evidence closes them.
Facta-Dev/ai-coding-playbook · ★ 0 · AI & Automation · score 65
Install: claude install-skill Facta-Dev/ai-coding-playbook
# <app> Claim Check skill A workflow that forces a deliverable about `<app>` to earn its claims. An agent invokes this skill before signing off on a plan, brief, report or recommendation; the skill supplies the evidence discipline, the agent supplies the claims. The goal is not more words, it is claims a reader can trace. ## When to Activate - An `<app>` deliverable (plan, brief, report, README, PR description) states something a reader will act on. - The request asks for a verdict, recommendation, or status that needs evidence behind it. - A claim looks confident but no source is visible next to it. Do not activate for a code change whose diff is itself the evidence, for pure transcription of an existing source, or for a claim the requester has already verified and cited. ## Core Principles - Every substantive claim carries an evidence id that resolves to a real source. - Confidence is tiered: observed (seen directly), derived (computed from observed facts), inferred (a judgement not directly witnessed). - Unknowns live in an open-questions ledger and close only when a cited id answers them. - A reader should be able to follow every claim to its source without asking a second time. ## Rules - No claim without an id: a claim that cannot point at a resolvable source is either downgraded to a question or dropped. - Never fabricate an id or point it at a source you have not read yourself. - Label the tier honestly: a number derived from two measurements is