More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Verify Before You Call It Done

    1

    Stop claiming work is done on confidence alone: name the proof for each claim, run it fresh, read all the output, and report only what the evidence supports.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Verify Before You Call It Done

    Verify Before You Call It Done

    Example session with this skill installed

    I just fixed a checkout bug in our online shop and I am about to tell the owner it is ready. The fix changes how discount codes apply at checkout. I ran the test suite this morning before my last edit, and my teammate says the payment flow looks fine. Tell me if I can call it done.

    • Read your context and instructions
    • Compiled the verify before you

    VERIFICATION RECORD — Checkout discount fix — state: current copy, after the final edit at 2:40 pm. Claims split before any check ran: (1) the discount bug is fixed; (2) nothing else broke; (3) the payment flow still works. The morning test run is STALE — it predates the last edit, so it proves nothing about the current work (evidence level 2). Your teammate's word is a handoff, not evidence (level 0). Fresh checks run: the exact failing case replayed — a 20% code on a two-item basket — now totals correctly, and a code applied twice is rejected on the second attempt. Full suite: 38 passing, 1 failing. The failure is in the receipts module, untouched by this fix: a currency symbol expectation left over from an earlier update. Payment flow checked directly on the real path: a test purchase completes and the receipt amount matches. VERDICTS — Bug fixed: VERIFIED (level 4, replay case + suite). Nothing else broke: PARTLY VERIFIED — suite is 38 of 39, with the one failure filed separately as unrelated. Payment flow: VERIFIED for the main path checked; the refund path was not exercised and is recorded as UNVERIFIED. Do not tell the owner "all done." Say: "Discount fix verified; suite 38 of 39 with one unrelated stale failure filed; refund path still unchecked." That sentence is the gate working.

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

    About this skill

    A completion claim is a statement about reality, and this skill treats it that way. It gives you one gate with no exceptions, applied before committing, pushing, opening a review, or telling anyone a task is done, fixed, or passing: name the exact check that proves each claim, run it fresh against the current state of the work, read the whole output including warnings and failure counts, and claim only what the evidence supports, with the evidence attached. A claim-to-evidence table settles the most common failure — a passing check that proves a different claim — covering tests, builds, bug fixes, requirement checks, delegated work, and artifacts. The freshness rule explains why a run from before the last edit is history, not proof, and why re-running after no change adds nothing. Language discipline catches the phrases that leak unverified claims ("should work now," "probably fixed," "just this once"), and a separate section covers verifying work done by others from artifacts rather than summaries, plus what to do when a check is genuinely blocked. The skill includes a copy-ready verification record template, a six-level evidence rubric from "none" to "fresh, full, independent," edge cases including flaky checks and lying environments, anti-patterns, and an eight-item pre-delivery quality gate. Every verdict is one of three words — verified, partly verified, or unverified — so the next person always knows exactly where things stand.

    - It disciplines verification; it does not run your tests, builds, or checks for you, and it cannot verify a claim when you have no way to run the proving check — in that case it records the gap as blocked instead. - It does not judge whether the work itself was the right work; a perfectly verified wrong feature still needs a requirements conversation. - Evidence levels depend on honest reporting of what ran and when; the skill gives you the language to be precise, not a way to detect a misreported run.

    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

    No. It sits on top of testing: it makes sure the right check actually ran, fresh, and that the claim matches what the output said. A team with no tests still needs to write them; this skill stops a team with tests from citing the wrong run.

    Then the claim becomes "verified except X, blocked by Y," with the blocker and the unblock step named. A blocked check reported as blocked is honest; one silently treated as passing is how untested work ships.

    The gate is four steps and usually takes minutes. If it feels slow, the checks are slow — fix them, or run the fast subset fresh and label the slow remainder explicitly. Tiny changes break things at the same rate as large ones.

    New in Productivity