メニュー
Execute one explicit verification command and record a minimal hash-bound AET proof receipt. Use only when the user explicitly invokes /aet-proof or asks AET to prove a command ran.
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
Create a bounded, read-only investigation from normalized Codex or Claude Code Run Records, optionally inspect explicit AET Proof and Freshness records, and export a Portable Evidence Bundle for an independent reviewer.
Run a bounded AET check of Agent instructions, Skills, verification contracts, and dangerous authority. Use only when the user explicitly invokes /aet-check or asks for this AET Quick check.
Deterministically check whether an AET proof still applies to the current workspace. Use only for an explicit /aet-fresh request.
Investigate whether an AI coding change matches the user's task and whether cross-module changes are necessary. Use only for an explicit /aet-scope request.
Compatibility entrypoint for AET Quick and advanced AET Lab workflows. Prefer the four dedicated Quick Skills for daily work; use this Skill only when the user explicitly asks for legacy CLI or Lab capabilities.
Held-out fixture with a mismatched directory name.
| name | aet-proof |
| description | Execute one explicit verification command and record a minimal hash-bound AET proof receipt. Use only when the user explicitly invokes /aet-proof or asks AET to prove a command ran. |
Answer one question: did this exact verification command run on this workspace?
Run:
aet quick proof --output <proof.json> \
[--relevant-path <path>] [--artifact <path>] -- <argv>
The user's explicit /aet-proof call authorizes writing the requested JSON
receipt. It does not authorize any other write. The host LLM may locate the
project's recommended command, smallest relevant test set, relevant paths, and
declared artifacts. The deterministic executor alone records argv, timestamps,
exit code, redacted log digests, Git/worktree binding, relevant file hashes,
environment binding, and artifacts.
Never describe an unexecuted or non-zero command as passed. Never reuse a historical result as a new execution. State that a bounded test does not prove the full test suite passed. Do not generate HTML, SVG, Viewer, or Evidence Pack.
Budget: command duration, at most one LLM call only when locating the command, zero remote calls by default, and one receipt write.
For a Chinese /aet-proof request, explain the receipt in natural Simplified
Chinese while keeping code and required technical terms in English. Otherwise
use English.
Write the one receipt, report its scope, and stop. Do not run additional tests or invoke Fresh unless the user explicitly requests them.