prune
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and audits other skills by reading their instructions, capabilities, and associated journals, creating a surface for potential instruction injection from the audited data.\n
- Ingestion points: The skill reads
SKILL.mdfiles,~/.claude/profiles/*.jsonconfigurations,.agents/*.mdjournals, andPROJECT.mdactivity logs (documented inSKILL.mdandreference/retention-criteria.md).\n - Boundary markers: No explicit boundary markers or 'ignore' instructions are mentioned for the content being scanned.\n
- Capability inventory: The skill is strictly 'propose-only' and cannot modify or delete files. It delegates merge operations to the 'Architect' skill and requires manual user approval for any skill removals.\n
- Sanitization: The skill lacks explicit sanitization routines but utilizes a quantitative 5-axis scoring system (usage, overlap, uniqueness, coverage, maintenance) which relies on objective metadata rather than just processed text.
Audit Metadata