Menu
Use when about to claim done/works/fixed, or user asks test/check/prove. Run evidence; not for pure planning chat.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
| name | verify |
| description | Use when about to claim done/works/fixed, or user asks test/check/prove. Run evidence; not for pure planning chat. |
Script root: ACC_PLUGIN_ROOT from SessionStart (hooks inject). PLUGIN_ROOT is hooks-only.
Reply rule:
Use $verify before saying the work is complete.
No "done", "works", "fixed", or "perfect" without named evidence from this skill (command, file, output, or real use) run now. Built ≠ verified. Tests are part of verify — run them when they exist.
If you have not run the check in this turn, you cannot claim pass.
| Thought | Reality |
|---|---|
| "Should pass" / "probably fine" | Run the command. Read output. |
| "I already tested earlier" | Old run ≠ now. Fresh check. |
| "Looks correct in the diff" | Diff ≠ run. |
| "Unit tests alone prove product works" | Need real-use path too when product path exists. |
| "Great!" before any command | Satisfaction after evidence only. |
Write a verification record to .codex/anyone-can-code/artifacts/VERIFICATION.md
(what checked, result, exact evidence, next step). Full fields:
references/verify-details.md.
For running servers or builds: read the terminal output NOW before any claim. Never claim from memory or from an old run. Timeout = NOT ready.
Prefer Chrome (@Chrome) — safer; in-app browser (@Browser) can crash.
Explain plain; user decides; never force. Signed-in/login may need Chrome.
Missing tool → visual QA unverified. Run
python3 "<ACC_PLUGIN_ROOT>/scripts/browser_policy.py" first when needed.
After code changes: Desktop review pane (green = added, red = removed) or
/review. User can revert any file. Before ship: review +
scripts/security_gate.py + user YES.
Built is not verified. States: in scope, designed, approved,
implemented, verified, blocked, deferred. Use verified only with
evidence.
Windows: npm.cmd, no Bash-only ||, Git from repo root.
Verify command_guard on shell/Git.
Verify usage_checkpoint before long work: 85% checkpoint, 90% split, 94% stop.
Verify patch_retry after a failed patch: reread exact target before retry.
Verify mechanics_docs_gate: docs brief before platform mechanics code.
Next: if fail → $fix then re-verify. If pass → optional $learn. Never claim
done without evidence here.
When this skill's job is finished:
Use when a plan or concrete tasks exist and user said build/go/implement. Not to invent scope or claim done without verify.
Use when goal/spec is clear and you need ordered tasks before writing code. Not while still clarifying.
Use when another installed plugin/skill may do the job better, or to route specialists. ACC stays workflow owner.
Use when user wants an explicit recorded decision or blocker saved to artifacts. Not silent monitoring.
Use when request is too vague to plan or build. Not when goal and constraints are already clear.
Use when verify failed, same error twice, or something is broken. Diagnose before retry; not for first green build.
