← ClaudeAtlas

double-checklisted

Exhaustively validate every claim, number, calculation, code-fact, and command in a document or outbound communication. Extracts all claims into a prioritized ledger (flagging what is NOT covered), validates each by the right method (run the math, run the command, cite the file:line, quote the primary source), then calibrates certainty with CIA Words of Estimative Probability and a provenance tier — and proposes corrected, properly-hedged rewrites for any overclaim. Use before sending a message, filing an issue, committing analysis, or whenever you state exact numbers/assertions others will rely on.
DazzleML/dazzle-claude-code-config · ★ 0 · AI & Automation · score 70
Install: claude install-skill DazzleML/dazzle-claude-code-config
# DOUBLE-CHECK — Exhaustive Claim Validation *(working title during design: CHECK-THE-NUMBERS)* ## Target: "$ARGUMENTS" Validate the claims in the target (a file path, a draft message, a section of analysis, or pasted text). The goal is to ensure we **are not overstating, understating, or fibbing** — and to say *how true* each claim is in calibrated terms. ## Three operating beliefs (the whole skill in one breath) 1. **A claim is only as strong as its provenance.** "I'm pretty sure" is not a source. 2. **Completeness must be proven, not implied.** If we didn't check something, we say so — out loud, in a list. 3. **Certainty must be stated, not assumed.** Every judgment gets a calibrated probability word, never a bare "possible" or unqualified "may." ## When to run it - Before sending an external communication (issue reply, email, PR comment, docs). - Before relying on your own analysis to make a decision (internal mode). - Any time the text contains EXACT numbers, registry/config values, code behavior claims, shell commands, or definitive factual assertions. - As the Ground-Truth engine for a `/dev-workflow-process`. ## Modes - **internal** — validating our own work before we trust it. Emphasis: catch unverified numbers and inference-as-fact. - **external** — validating something we will send. Everything in internal, PLUS grade tone/overclaim and propose hedged rewrites the recipient will read. Default to **external** when the target is addressed to anyone but us. ---