runlisted

# APE run Use this skill only when the user explicitly asks to run APE. Do not infer consent from a coding request or from plugin availability. Before `ape_run start`, inspect the repository and compose bounded, cold-reader-complete inputs: When the host is Google Antigravity / Gemini, pass the exact open project root as `project_dir` on this and every later APE MCP call. - `objective`: the observable outcome and acceptance criteria. - `mode`: `phase`, `debug`, `spike`, or `land` for an already-finished diff. - `lane`: normally `auto`; `mechanical` is non-behavioral docs/config/generated work, `fast` is a bounded behavioral change, and `full` is planned multi-stage work. - `claimed_paths`: production paths only. Include generated artifacts or obvious documentation surfaces only when the objective may require them. - `test_paths`: independently authored test paths for behavioral work; never put them in `claimed_paths`. - `behavioral`, `requirements`, `completes`, `risk_triggers`, and least-privilege `tool_claims`. - `plan_contract_version: 2` for every newly started behavioral fast or full `phase` run. Omit it for mechanical and non-phase modes, and every resume; explicit version 1 is only for legacy compatibility. Do not invent missing load-bearing product choices. Recommend concrete options and ask only for decisions that materially change the outcome. If the work clearly spans multiple runs, offer a roadmap; registration still requires explicit approval. Confi