agent-engineering-toolkitlisted

# Agent Engineering Toolkit Current Skill version: `1.12.0` (Evidence → Quality → conditional Gate → bounded Evolution) ## Activation policy **Default: OFF.** Installing this Skill does not authorize its use. Load or run AET only when the user explicitly asks to use AET for the current task (for example, “use AET”, “run `aet audit`”, or “produce an AET Evidence Pack”). A repository containing AET files, a generic request to test/review a change, or the availability of the `aet` executable is not opt-in. Do not carry opt-in across tasks. After explicit activation, choose only the smallest surface needed. Prefer `aet evidence receipt` when a compact status is sufficient. A successful Trace may be reused only through explicit `aet trace --reuse-if-fresh`, which refuses command, proof, artifact, log, or workspace drift without executing anything. Do not run real-host replay, Gate, tournament, or Sleep unless the user separately asks for governance-asset evaluation or evolution. Do not repeat a proof command that already has fresh, hash-bound evidence for the unchanged workspace. Use the `aet` CLI as the source of truth. The host agent may choose its own shell or package runner, but must preserve the commands' exit status and attach the emitted evidence instead of paraphrasing it as unverified fact. <!-- aet-learn:immutable --> `UNKNOWN` is never a pass. Only `aet trace` executes explicit argv after `--`. Audit, review, and Evidence Pack compilation stay deterministic and lo