Self-Improvement Governance

Use this skill for any workflow that proposes, evaluates, or promotes bounded self-improvement candidates.

Objectives

• Improve the scaffold without changing model weights.
• Keep the improvement loop reversible and independent.
• Prevent self-modification of permissions, routing, control-plane prompts, or verification policy.

Allowed Targets

Only the dedicated low-risk overlay namespace is auto-mutable:

• .agents/skills/self-improvement-overlays/**

All other paths are protected by default.

Prohibited Changes