mirror of
https://github.com/EveryInc/compound-engineering-plugin.git
synced 2026-09-19 01:09:55 +08:00
Trevin Chow
96c06e3dfc
The previous revision justified the autonomy directive by pointing at the harness's reserve-blocking-questions guidance, which was borrowing the authorization directive's credibility. That instruction pushes toward inferring, but it asserts nothing about user presence, so it does not satisfy the directive's own trigger condition. Checked directly: no autonomy assertion appears in either an interactive or a headless Claude Code system prompt, and the directive self-limits to nothing where none exists. Its real basis is operational — long-running work drifting toward inferring instead of asking — and it is carried as cross-host insurance for agents and schedulers where such framing is plausible. Labelled as such rather than dressed up as measured. Claude-Session: https://claude.ai/code/session_016GD4nZMi43krCgTT5kpjsx