More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Fact Check: Claims Against the Code, With Evidence

    1

    Verify a design doc, spec, or plan claim-by-claim against the actual codebase — every API name, path, count, and mechanism checked with evidence and a verdict.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Fact Check: Claims Against the Code, With Evidence

    Fact Check: Claims Against the Code, With Evidence

    Example session with this skill installed

    Check this claim from our auth redesign doc: "The auth middleware runs before the router, we have 47 endpoints, and sessions are cached with a database trigger for expiry."

    • Read your context and instructions
    • Compiled the fact check

    Auth redesign claims — checked with the skill

    Three claims extracted verbatim, three verdicts with evidence:

    1. "The auth middleware runs before the router" — CONFIRMED. Evidence: src/server.ts:88 — app.use(authMiddleware) precedes app.use(router) at line 94.
    2. "We have 47 endpoints" — REFUTED. Evidence: grep -c across the route files counts 52 endpoints (method shown in the report). Correction direction: the migration estimate was built on 47; re-estimate on 52.
    3. "Sessions are cached with a database trigger for expiry" — REFUTED twice over. There is no cache layer in the code, only an in-memory map with no expiry — and the codebase has no database at all, so a database trigger cannot exist. Correction direction: the session design needs a real expiry mechanism; the trigger was the plan's core mechanism and it cannot exist.

    Tally: 1 confirmed, 2 refuted, 0 unverifiable. Mode: BLOCKED — the refuted claims need doc or code updates before this plan is approved. The doc's owner decides whether the doc or the code changes; this report does not rewrite anything. Rubric score: 9/10 — one point held back because the endpoint count depends on whether health-check routes count; the report states its definition and notes the alternative count.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Design documents drift from code. Plans cite APIs that were renamed, paths that moved, and mechanisms that were never possible. This skill is the lightweight correction: extract every verifiable claim from a document, quote it verbatim, and check each one against the codebase with evidence — file, line, and excerpt, or the counting method for numbers. Verdicts are confirmed, refuted, or unverifiable, and refuted claims carry the correction direction (what reality is), not a fix. The summary tally states the mode plainly: refutations block approval until the doc or the code changes, the owner's choice.

    You get a claim-extraction template and a verdict template, a checklist, and a scoring rubric with a ship threshold of 8/10. The boundaries are explicit: not a logic check, not a security audit, not a design review — a checked claim is a true claim, not a safe or well-designed one. Edge cases cover the claim that moved, the count that depends on its definition, the mechanism whose parts exist but never connect, and the future state that cannot be checked. A worked example runs an auth-redesign doc through the full check: 23 claims, 18 confirmed, 3 refuted, 2 unverifiable — and the plan does not proceed on a mechanism that cannot exist.

    • It is not a logic check — state machines, retry protocols, and "always/never" claims need executable verification.
    • It is not a security audit — a true claim is not a safe claim.
    • It is not a design review — "the API exists" and "the API is well-designed" are different verdicts.
    • It cannot check claims about external services or future states; those are marked unverifiable with the reason.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Because paraphrasing a claim before checking it is how you accidentally check a different claim — and the paraphrase is usually the easier, truer version, which defeats the purpose. The verbatim quote is the check.

    The report states what reality is — the correction direction — and the mode becomes blocked. The doc's owner decides whether the doc or the code changes. The report never rewrites the document itself.

    Claims that cannot be checked against the codebase: external dependencies, future states, judgment calls. They are marked with the reason and, where possible, the skill that could settle them — instead of getting a confident verdict they cannot support.