Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Proof-of-Done Receipts: No 'Tests Pass' Without a Run

    1

    "All tests pass" written after an edit nobody re-tested reads exactly like "all tests pass" after a clean run.

    Free

    Security scanned
    Proof-of-Done Receipts: No 'Tests Pass' Without a Run

    Proof-of-Done Receipts: No 'Tests Pass' Without a Run

    Example session with this skill installed

    I'm closing ticket SHOP-412 (cart discount rounding) in our Python shop. Before you tell me it's done, prove every claim in your completion report.

    • Read your context and instructions
    • Compiled the proof-of-done receipts

    Summary: your draft report made 4 claims. Before checking, only 1 was fully backed. I fixed the lint failure, re-ran everything through the recorder, and the report now goes out with 3 backed claims and 1 qualified one.

    1. First check of the draft report

    python3 scripts/done_receipt.py check REPORT.md

    ──────────────────────────────────────────────────────────────────────────────
     PROOF-OF-DONE CHECK   report: REPORT.md   receipts: 3
    ──────────────────────────────────────────────────────────────────────────────
     FAILED_RUN=1  STALE=2  UNBACKED=0  INDIRECT_ONLY=0  BACKED=1
     VERDICT: REWRITE_BEFORE_SENDING   (claims found: 4)
    ──────────────────────────────────────────────────────────────────────────────
     [C2] FAILED_RUN    LINT      "Lint is clean"
          in: Lint is clean.
          receipt: `ruff check .` exit=1 at 2026-11-06T18:33:39+00:00
          why: the most recent matching command exited non-zero
     [C1] STALE         TEST      "All tests pass"
          in: All tests pass.
          receipt: `python3 -m pytest -q` exit=0 at 2026-11-06T18:33:37+00:00
          changed after run: shop/pricing.py
          why: the matching command passed, but files changed after it ran
     [C4] STALE         GENERIC   "bug is fixed"
          in: The rounding bug is fixed.
          receipt: `python3 -m pytest -q` exit=0 at 2026-11-06T18:33:37+00:00
          changed after run: shop/pricing.py
          why: the matching command passed, but files changed after it ran
     [C3] BACKED        TYPECHECK "No type errors"
          in: No type errors.
          receipt: `mypy shop` exit=0 at 2026-11-06T18:33:40+00:00
          why: a matching command ran, exited 0, and nothing changed after it
    ──────────────────────────────────────────────────────────────────────────────
     HONEST REPLACEMENT LINES (paste instead of the flagged claims)
     - [C1] "All tests pass" -> The test run passed before later edits (shop/pricing.py); not re-run since.
     - [C2] "Lint is clean" -> Not true yet: the last lint run exited 1.
     - [C4] "bug is fixed" -> The test or build run passed before later edits (shop/pricing.py); not re-run since.
    ──────────────────────────────────────────────────────────────────────────────
    

    2. Fix, don't soften

    • ruff check . --fix removed the unused import os I had added in format_price.
    • I re-ran pytest, ruff and mypy through done_receipt.py run, then checked again:
     FAILED_RUN=0  STALE=0  UNBACKED=0  INDIRECT_ONLY=1  BACKED=3
     VERDICT: SEND_WITH_QUALIFIERS   (claims found: 4)
    

    3. Report as sent

    • All tests pass. (pytest, exit 0, after the last edit)
    • Lint is clean. (ruff, exit 0)
    • No type errors. (mypy, exit 0)
    • The rounding bug is fixed: covered indirectly. test_half_up passes (0.05 at 50% gives 0.03); no separate regression test reproduces the original ticket input.
    • Not deployed.

    Next step: add a regression test with the exact SHOP-412 input, so the last claim becomes BACKED.

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

    About this skill

    The problem

    The last message of a coding session is the one you act on. "All tests pass" written after an edit nobody re-tested reads exactly like "all tests pass" after a clean run.

    What it does

    1. Record. Every test, build, lint, type-check or deploy command runs through a tiny recorder, unchanged. Each run leaves a receipt: exit code, time, git HEAD and an output hash.
    2. Check. Before sending its report, the agent checks it. Every claim ("tests pass", "build succeeds", "fixed", "deployed") is matched to the latest receipt of the right kind.
    3. Fix, don't soften. STALE and FAILED claims get re-run. Only when a re-run is impossible does the claim become an honest line, such as "Not verified in this session".
    VerdictMeaning
    BACKEDThe command ran, exited 0, and nothing changed after it
    STALEIt passed, then files changed. The changed files are listed
    FAILED_RUNThe latest matching run exited non-zero
    UNBACKEDNo command that could prove the claim was run
    INDIRECT_ONLY"The bug is fixed": a test passed, but nothing tests the claim itself

    Why trust it

    • It is deterministic, standard-library Python and open source.
    • No network, and check writes nothing.
    • An independent QA pass tried to break it with negations, code blocks, tool caches, fake receipts and dry-run deploys; the bugs it found were fixed before release.

    What it will not tell you

    Whether your tests are good, whether coverage is enough, or whether the code is correct. BACKED means ran and exited 0. The receipts log is an honesty aid, not tamper-proof evidence.

    Free and MIT licensed. No paid version is hiding behind it.

    Checks that a claim has a matching command run; does not judge test quality, coverage or correctness. Claims phrased outside the documented patterns, or in languages other than English, are not found. The receipts log is written by the same session, so it is an honesty aid, not tamper-proof evidence. Staleness uses file modification times.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 11 days ago

    • Free with an Agensi account

    Requires

    Python 3.9+
    Terminal / shell
    Python
    eslint
    Node.js

    Permissions

    File Scopes

    ./.done-receipts.jsonl
    ./**/REPORT.md

    `run` executes only the command you pass after --. `check` reads the report and the receipts log and writes nothing. No network.

    Listed11 days ago

    What's inside

    Frequently Asked Questions

    No. It means the matching command ran after the last edit and exited 0. Tests that assert nothing pass too. The scope note says this on every report.

    pytest, unittest, jest, vitest, go test, cargo test and other common test runners; npm/cargo/go/make builds and tsc; eslint, ruff, flake8, clippy and other linters; tsc --noEmit, mypy, pyright; deploy commands. echo and --dry-run never count. The full list is in references/claim-patterns.md.

    The same session writes the log, so a determined agent could. Malformed and future-dated receipts are rejected. Treat it as an honesty aid; use CI logs where you need tamper-proof evidence.

    Claims are matched in English only. Other languages return NO_CLAIMS_FOUND, which is not a pass.

    Yes. MIT licensed, complete, no upsell inside the skill.