The Agent Skills Directory

[SAFE]: The skill implements a structured reflection protocol that identifies instructional weaknesses such as ambiguity or unhandled edge cases based on actual execution signals.
[SAFE]: Mandatory human-in-the-loop verification is enforced. The protocol explicitly prohibits automatic application of updates (rule:no-auto-apply), requiring all proposed diffs to be reviewed and approved via the /ask-gate mechanism.
[SAFE]: Recursion protection is integrated into the workflow. The skill uses a depth-tracking marker (SIP-RESOLVED) to prevent infinite loops that could occur when the self-improvement logic triggers on its own reporting or gating tools.
[SAFE]: No unauthorized data access or network exfiltration patterns were detected. The skill relies on internal agent reflections and local file analysis without external dependencies or remote script execution.

self-improve