skill-stocktake
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Executes included bash scripts (scan.sh, quick-diff.sh, save-results.sh) to automate file inventory, comparison, and result merging. Scripts use standard system utilities and implement safety checks like path validation and random temporary file suffixes.
- [DATA_EXFILTRATION]: Reads local skill files (SKILL.md) and usage observation logs (~/.claude/observations.jsonl) to gather metadata and frequency data for the audit report. All data processing is contained within the local environment.
- [PROMPT_INJECTION]: As an auditing tool, the skill extracts metadata (name, description) from other skills to perform a quality evaluation. This represents an indirect prompt injection surface where untrusted content in the audited files could attempt to influence the evaluation verdict. Ingestion points: scan.sh (metadata extraction from audited files). Boundary markers: Absent. Capability inventory: scan.sh, quick-diff.sh (read), save-results.sh (write). Sanitization: Absent.
Audit Metadata