Back to Neuro OS Playbook
Feedback Loop Skill
Source: playbook-source-package/skills/feedback-loop/SKILL.md
Feedback Loop
A mandatory skill for closing the loop after every Founder correction. Ensures mistakes are logged, instructions updated, and fixes verified.
Description
When the Founder points out a mistake, this skill ensures the correction is permanently captured in agent instructions, preventing the same error from recurring. Without this, corrections are acknowledged but not embedded — leading to repeated failures.
Procedure
After every Founder correction (MANDATORY)
- Identify root cause. Why did this happen?
- Update the relevant agent's HEARTBEAT.md. Add the failure to their anti-patterns table.
- Update the relevant agent's AGENTS.md if the Definition of Done needs a new criterion.
- Update PROJECT-INVENTORY.md if deliverables changed.
- Verify the fix. On the next similar task, check that the same mistake does not recur.
- Log it. Daily notes: what went wrong, what was updated, what to watch for.
Severity levels
- Once: A bug. Fix it, log it.
- Twice: A process failure. The instructions need updating.
- Three times: The instructions are not clear enough. Rewrite them entirely.
Applicable Roles
- CEO
- All agents (as recipients of instruction updates)
Tags
- feedback
- learning
- anti-patterns
- mandatory