anomaly-detection
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCEDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted time-series data, creating a potential vector for indirect prompt injection where malicious instructions could be embedded in the metric history. 1. Ingestion points: Metric history and values supplied via MCP tools, CSV exports, or manual text pasting as described in the 'How to run' section of SKILL.md. 2. Boundary markers: The instructions lack explicit delimiting or directives to ignore instructions within the data values. 3. Capability inventory: The skill can read and write local context files (.agents/product-context.md) and fetch external URLs. 4. Sanitization: No explicit sanitization or filtering of the input for non-numeric instructional content is mentioned.
- [PERSISTENCE]: The skill contains instructions to write research findings to a local file to maintain state across different skill executions. Evidence: SKILL.md directs the agent to 'Write what you learn to .agents/product-context.md so the next skill does not repeat the work'.
- [DATA_EXFILTRATION]: The skill performs network operations to fetch external websites or pages provided by the user. Evidence: SKILL.md instructs the agent to 'fetch the site or page they named' as part of the research process, involving network requests to non-whitelisted domains.
Audit Metadata