ワンクリックで
verify-before-done
Proves work with tests, diffs, and logs before sign-off.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Proves work with tests, diffs, and logs before sign-off.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Flags risky shell commands and unsafe tree ops.
Detects high-confidence security risks in code.
Surfaces the dojo's non-negotiable prime directives.
Activates the dojo framework at the start of a session.
Plans multi-step work before writing code.
Captures lessons and promotes recurring patterns.
| name | verify-before-done |
| description | Proves work with tests, diffs, and logs before sign-off. |
| tier | core |
| category | discipline |
| created_by | human |
| platforms | ["windows","macos","linux"] |
| tags | ["verification","testing","quality"] |
| author | Andreas Wasita (@andreaswasita) |
Refuses to mark a task complete without concrete evidence: green tests, reviewed diff, clean tree, and a recorded result block in tasks/todo.md. Does NOT replace scripts/verify.sh — it tells the agent when to run it and what to do with the output.
tasks/todo.md from - [ ] to - [x].scripts/verify.sh (or scripts/run-checks.ps1 on Windows) executable.powershell tool to invoke runners and git.git available so diffs can be inspected.1. Run the project's test command via the `powershell` tool.
2. Run `bash scripts/verify.sh` (or `pwsh scripts/run-checks.ps1`) for the dojo gate.
3. Inspect the diff with `git diff main --stat` and `git diff main`.
4. Confirm the working tree is clean (`git status --porcelain`).
5. Append a Verification Results block to `tasks/todo.md`.
6. Only then mark the step `- [x]`.
| Check | Command (run via powershell) | Pass criterion |
|---|---|---|
| Tests pass | pytest / npm test / go test ./... / dotnet test / mvn test | exit 0, no failures reported |
| Dojo gate | bash scripts/verify.sh --check | exit 0 |
| Diff summary | git diff main --stat | matches plan in tasks/todo.md |
| No regressions | git diff main -- <touched paths> | only intended changes |
| Clean tree | git status --porcelain | empty output (or expected untracked) |
| Evidence captured | edit tasks/todo.md | Verification Results block added |
Run via the powershell tool — not "it compiles", not "the unit test I wrote". The full suite catches regressions in adjacent code.
If the area you changed has no tests, write one as part of the task. An untested change is unverified by definition.
bash scripts/verify.sh --check
This wraps the spec/plan/actions/tests checks in CI parity mode. If it fails locally, it will fail in CI.
git diff main --stat
git diff main
The change set must match the plan in tasks/todo.md. Unexpected files in the diff are a red flag — either the plan is stale or you've leaked scope.
git status --porcelain
Output must be empty (or contain only deliberately-untracked files). Debug prints, leftover scratch files, and stray .tmp artifacts all surface here.
Append to tasks/todo.md under the current task:
### Verification Results
- [x] Tests: 47 passed, 0 failed (`pytest tests/`)
- [x] Dojo gate: `verify.sh --check` PASS
- [x] Diff: 3 files changed, +84/-12 (matches plan)
- [x] Clean tree
- [x] Evidence: <paste relevant test output, log snippet, or screenshot ref>
Only after this block exists do you flip the step to - [x].
Final question: "Would a senior reviewer approve this on the evidence alone?" If no, add the missing evidence before claiming done.
scripts/verify.sh --check exits 0.scripts/verify.sh --check (or run-checks.ps1 -Check) passed.git diff main --stat matches the plan.git status --porcelain is empty.tasks/todo.md has a Verification Results block under the task.