The Agent Skills Directory

[SAFE]: The skill adheres to a strict 'propose-not-apply' security posture for constitutional files such as CLAUDE.md and MEMORY.md. It generates unified diffs for conductor review and manual application, preventing unauthorized modifications to critical system documentation.\n- [COMMAND_EXECUTION]: Utilizes local utility scripts (audit-citations.sh, find-orphan-plans.sh, propose-citation-fix.py) to perform filesystem audits and documentation maintenance. These scripts are self-contained and perform scoped read operations or in-memory string processing without executing external payloads.\n- [SAFE]: Indirect Prompt Injection Surface: The skill processes data from CLAUDE.md, MEMORY.md, and plan files (~/.claude/plans/*.md). While ingestion of this content occurs, the skill maintains security through its 'propose-not-apply' protocol and limits capabilities to local script execution and metadata logging. Sanitization is performed via shell escaping in the audit script and standard library handling in the proposer script.

artifact-resurfacing