What problem does it solve? Executor skills often stall mid-task by asking unnecessary confirmation questions, breaking autonomous workflows. This protocol establishes a single shared rulebook so executor skills run uninterrupted from plan approval to final report, stopping only at three defined checkpoints. ## Core Features & Use Cases - Three Stop Points: Execution halts only for plan approval, merge/commit to master, or production deploy; all other questions are forbidden. - Pre-flight Memory Check: Requires scanning project MEMORY.md for feedback and project lessons before the first non-trivial tool call. - Uncertainty Resolution Policy: Ordered decision rules resolve ambiguity without asking the user, with all decisions logged in the final report. - Self-Improvement Rule: After any error, the skill must update its own SKILL.md or LESSONS.md so the mistake cannot recur. - Use Case: A teamlead skill delegates a backend task; the backend executor imports this protocol, inherits plan approval, and runs through implementation, review iterations, and feature-branch commits without asking a single intermediate question. ## Quick Start Reference this protocol from an executor skill by adding a line stating that the autonomy-protocol applies, then let the skill execute the approved plan autonomously.
