skill-monitor
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes session metrics from local files and passes them to a subagent for qualitative analysis, which introduces a theoretical surface for indirect prompt injection. 1. Ingestion points: The skill reads session metrics from the '.claude/session-metrics/metrics.jsonl' file as described in SKILL.md Step 2. 2. Boundary markers: The prompt provided to the 'skill-effectiveness-analyzer' agent in Step 5 lacks explicit delimiters or instructions to treat the interpolated metrics data as untrusted content. 3. Capability inventory: The skill utilizes file read/write operations within the '.claude/' directory and subagent spawning capabilities. 4. Sanitization: The instructions do not define specific sanitization or validation logic for the ingested JSONL data before prompt interpolation.
- [SAFE]: No malicious patterns such as credential harvesting, unauthorized network communication, or persistent access mechanisms were identified. The skill's behavior is consistent with its stated purpose of monitoring and improving agent effectiveness.
Audit Metadata