Fixes from functional correctness review: - PHASE_PRIORITY: add code_review (was missing, caused wrong task selection) - cmd_drive: implement→code_review (was illegal implement→bug_find) - cmd_drive: add handlers for code_review and code_review:approved - _all_tasks: skip tasks/complete/ (was showing completed tasks as phase=None) - _all_tasks: recurse into subtasks/ (subtasks were invisible to autopilot) - _all_tasks: use .automaton/tasks for non-framework projects (was project/tasks) - is_terminal: only True for complete (human_intervention has legal transitions) - needs_user_input: flag human_intervention as blocked (was auto-driving it) Tests cover all 8 bugs: PHASE_PRIORITY ordering, every phase transition, code_review handlers, complete/ exclusion, subtask recursion, project path, terminal state, user-input detection, human_intervention handling.
- Fix decomposition:approved and human_intervention dead-end phases - Add scripts/autopilot.py: real drive_all() implementation - Fix guard plugin: throw Error instead of injecting user messages - Fix status.py: double continue, _require_state, --list-states - Add Category 3 (git-based modification) audit - Add Category 5 (stuck-task detection) audit - All 206 tests pass