Evidence before Done

    1

    Stops your coding agent from saying "done" without proof. Before it reports a task as finished, it must record what it checked and the result (test run, file, review, approval) in a file in your repo.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Evidence before Done

    Example session with this skill installed

    You fixed the login timeout. Before you tell me it is done, follow evidence-before-done for the task fix-login-timeout.

    • Read your context and instructions
    • Compiled the evidence before done

    Example session. The script output below is real; the project and its test numbers are illustrative.

    1. Check for existing proof

    $ node scripts/evidence.mjs status fix-login-timeout
    UNVERIFIED fix-login-timeout: no evidence recorded.
    

    2. A vague note is rejected

    $ node scripts/evidence.mjs add fix-login-timeout test-run --note "ok"
    Evidence needs a link or a note of at least 10 characters saying what was checked and the result.
    

    3. Run the real check, then record it

    $ node scripts/evidence.mjs add fix-login-timeout test-run --note "npm test: 42 tests, 42 passed; login completes in 1.8 s with timeout set to 5000 ms"
    Recorded test-run on fix-login-timeout. fix-login-timeout is VERIFIED (1 item(s)).
    

    4. Confirm and report

    $ node scripts/evidence.mjs status fix-login-timeout
    VERIFIED fix-login-timeout
    - test-run 2026-11-14  npm test: 42 tests, 42 passed; login completes in 1.8 s with timeout set to 5000 ms
    
    TaskStatusEvidence
    fix-login-timeoutVERIFIEDtest-run: 42 tests, 42 passed; login 1.8 s

    Report to the user: the login timeout fix is done and verified. Evidence: test run, 42 of 42 tests passed, login completes in 1.8 s. The record is in .evidence/evidence.json; commit it with the fix.

    Next steps: run node scripts/evidence.mjs list at any time to see which tasks have proof and which do not

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

    About this skill

    The problem

    Agents say "done" when a command exits without an error. That is not the same as the result being checked. You find out later that the tests were never run or the file was never opened.

    What this skill does

    It gives your agent one rule and one small tool:

    • Before reporting a task as done, fixed or shipped, the agent runs evidence.mjs status <task>.
    • If the task is UNVERIFIED, the agent has to run the real check (tests, build, open the file, read the reviewer's message) and record it with evidence.mjs add <task> <type> --note "what was checked and the result".
    • Only a VERIFIED task may be reported as done, and the agent quotes the evidence in its report.
    • evidence.mjs list audits every recorded task.

    Evidence types: test-run, file, review, customer-approval, payment-proof. Each item needs a link or a note of at least 10 characters. Records are stored in .evidence/evidence.json in your project, so you can commit and review them.

    What is inside

    • SKILL.md: when to use it, the steps and the rules (never invent a result; a check you did not run is not evidence; failed checks are reported, not recorded).
    • scripts/evidence.mjs: a single Node script with no dependencies.

    Limits

    • The skill records what the agent says it checked. It does not verify that a link or a note is true. It makes skipped checks visible; it does not make them impossible.
    • Needs Node 18 or newer.
    • New skill: no usage numbers yet.

    A Jira version with a workflow gate and a Rovo agent is open source at github.com/ahmetilhanli/evidence-gate.

    Needs Node 18+ and permission to run node and write .evidence/evidence.json. It records what the agent says it checked; it does not verify that a note or link is true.

    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 2 days ago

    Permissions

    Read Files
    Terminal / Shell
    Write Files
    Listed2 days ago

    What's inside

    Frequently Asked Questions

    It makes that much harder to do silently. The skill tells the agent to run a status check before reporting, and the script returns UNVERIFIED until evidence with a link or a real note is recorded. It is an instruction plus a record, not a hard technical lock: an agent that ignores its instructions could still skip it, and you would see the missing record.

    Any agent that reads SKILL.md skills, can run a shell command and has Node 18 or newer. It was written and tested with Claude Code. Other agents have not been tested by us yet.

    Yes. Evidence is stored as plain JSON in .evidence/evidence.json inside your project. You can read it, review it in a pull request and commit it with the work.

    No. The skill does not run your tests. The agent runs whatever check your project already has and records what it ran and the result.

    A SKILL.md with the steps and rules, and one dependency-free Node script (scripts/evidence.mjs) with three commands: add, status and list.

    Node 18 or newer, and an agent allowed to run node and write files in the project. Nothing else to install, no network access.

    New in LLM Ops