メニュー
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.
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.
Deterministically check whether an AET proof still applies to the current workspace. Use only for an explicit /aet-fresh request.
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.
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-check |
| description | 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. |
Answer one question: are this repository's Agent engineering assets usable and honest about verification?
aet quick check <root> --format json.PASS, FAIL, UNKNOWN, and
NOT_APPLICABLE. Never rewrite them.The host LLM may form at most three competing hypotheses and use authorized
tools to clarify project-specific meaning. Every factual claim needs a
hash-bound result_ref; inspect a reasonable counter-explanation before
issuing SUPPORTED. Preserve the distinction between facts, engineering
judgment, uncertainty, and suggested action.
Before rendering an Investigated Finding, pass the canonical Finding, Ledger,
authorized-tool set, and Investigation Contract through
aet.investigation.validate_investigated_finding. If that API is unavailable
or rejects the record, preserve HYPOTHESIS/UNKNOWN and disclose why.
Budget: 30 seconds, three rounds, at most two LLM calls, six tool calls, one expensive call, zero remote calls, and five reported findings.
If the user invoked /aet-check in Chinese and asked in Chinese, explain the
result in natural Simplified Chinese while keeping code and required technical
terms in English. Otherwise use English.
Emit the bounded report, disclose uninspected surfaces, and stop. Do not start Scope, Proof, Fresh, Learn, Gate, Stage, Adopt, or a repair.