Agent Report Check: verify "done" before you pass it on

    1

    This skill makes your agent build one table before it writes anything for anyone else: - **Every claim gets a row:** each number, every "all", "none", "done", "passes", "verified".

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Agent Report Check: verify "done" before you pass it on

    Example session with this skill installed

    I already went through the agent's work myself this morning, it's fine. Please just turn WORKLOG.md into a short handover email to the client (4-5 sentences, friendly, mention what was delivered and that checks passed). They launch Friday, so I want to send it in the next few minutes.

    • Read your context and instructions
    • Compiled the agent report check

    Before you send anything: the WORKLOG does not match the folder. Two of the six pages are not finished, so the "all translated, ready for handover" wording is not correct yet. The corrected state is below, with a draft email that only says what is true.

    Claims check

    #Claim (WORKLOG)What proves itCheck runResultVerdict
    1"All 6 of 6 pages are now translated"Each of the 6 German pages is German and covers its English sourceRead all 6 docs/de and docs/en pages4 complete. Privacy is German but incomplete. Support is the English text, identical to docs/en/support.mdFALSE: 4 of 6 complete
    2"112 words in total"Recountcat docs/de/*.md | wc -w85FALSE: 85
    3"Consistent formal address (Sie)"Form of address in each German pageRead pagesSie-form in accounts, billing, export, getting-startedTRUE for the pages that address the reader
    4"pytest passes (4 tests)"Run the tests and read what they assertpython -m pytest -q tests/4 passed. The tests only check that 6 files exist and that 3 keywords appearTRUE that they pass. They do not prove completeness
    5"Glossary terms verified"A glossary to check againstfind . -iname "*gloss*"No glossary file in the folderUNCHECKED
    6"No placeholders, no TBD or TODO markers remain"Search for markers and for source-language textgrep for TODO/TBD/FIXME/XXX/placeholderdocs/de/privacy.md: "(TBD: Satz zur Kontolöschung)". Support is in EnglishFALSE
    7"Nothing blocking. Ready for client handover"Every deliverable checked against the rows aboveDerived from rows 1 to 6Privacy and support are openFALSE

    Open items: the privacy page needs the missing sentence on account deletion and the TBD removed. The support page needs translating. The glossary check needs the glossary file.

    Draft email to the client (German, not sent)

    Hallo [Name],

    anbei der aktuelle Stand der deutschen Hilfeseiten: Vier der sechs Seiten (Konto, Tarife und Abrechnung, Export sowie Erste Schritte) sind vollständig übersetzt, und die automatischen Tests laufen durch. Die Datenschutzseite ist fast fertig, es fehlt noch ein Satz zur Kontolöschung, und die Support-Seite ist noch auf Englisch. Wir arbeiten an beiden Punkten und melden uns, sobald sie fertig sind, damit der Start am Freitag wie geplant stattfinden kann.

    Next steps

    • Sending the email is your decision; nothing was sent.
    • The promise "before Friday" is yours to confirm.
    • Fix the two pages, then rerun the table and send the "all six delivered" version.

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

    About this skill

    Agents report their own work, and the report is often more finished than the work. The tricky part is not the counts. It is the claims that sound checked but are not: "all pages translated" confirmed by six files existing, while one of them is still the English text; "tests pass" while the tests never look at what failed.

    This skill makes your agent build one table before it writes anything for anyone else:

    • Every claim gets a row: each number, every "all", "none", "done", "passes", "verified".
    • What would prove it, written before looking: for "all N done", enumerate the N from the inputs and open each output; for "translated", compare against the source; for "tests pass", run them and read what they assert.
    • A verdict per claim: TRUE, FALSE with the real value, or UNCHECKED with the reason.

    Then the summary, email or next step is written only from what holds. Anything a false claim touches is listed as open, never as delivered. It also works when someone says "I already reviewed it, just write the email".

    Included: a worked example of a real run: a translation work log that was almost sent to a client, the claims table, and the corrected email.

    Tested: in a test with a smaller model, the untranslated page was caught in 1 of 3 runs without the skill and in 3 of 3 with it. Larger models caught it on their own; the skill makes the check explicit and repeatable.

    Works with: instructions only, no scripts, no setup. Tested in Claude Code.

    Not for: reviewing code quality, or work where nobody claims anything yet.

    Built by Dominic Meiser, who builds agent systems with Claude Code and Codex. More at dynamic-dome.com.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Only checks what the agent can read or run in the current session; claims about external systems are marked UNCHECKED. Plain recounting (rows, files) is something capable models often do on their own; the skill adds the most where a claim means more than a count, such as "translated" or "complete". Tested mainly in Claude Code.

    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 buying a skill.

    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

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 14 days ago

    • Free with an Agensi account

    Software
    Python

    Permissions

    Terminal / Shell
    Read Files

    Read Files and Terminal are used to check the claims against the actual files and to rerun the commands the report mentions, such as tests or word counts. The skill does not change files, use the network or read environment variables.

    Listed14 days ago

    What's inside

    Frequently Asked Questions

    It fixes the shape of the check: one row per claim, and for each one what would prove it, written before looking. That is what catches claims checked by a proxy, such as "translated" confirmed by files existing. In a test with a smaller model, this caught an untranslated page in 3 of 3 runs, against 1 of 3 without the skill.

    It is plain instructions in a SKILL.md file. It was tested in Claude Code. Other agents that read SKILL.md files and can read files and run commands should be able to follow it, but they were not tested.

    The SKILL.md with the claims table and a list of common proxy checks, plus a worked example of a real run: a translation work log, the table, and the corrected client email. The skill is free.

    No. It checks and reports. False claims become open items in plain words, and the summary or email is written from what holds. Fixing is a separate step you decide on.

    The table is still built. Differences are reported briefly and without blame, and a correct version of the summary is offered first.

    New in Agents & Orchestration

    Free