optimize-mcp-tool-metadata
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates AI-driven experiments by ingesting external data from evaluation datasets (e.g., './data/host-evals.json') and using that data to prompt LLMs for tool-triggering analysis.
- Ingestion points: The skill reads input from a user-provided or generated dataset file via the
loadEvalDatasetfunction inSKILL.md. - Boundary markers: The instructions do not specify boundary markers or "ignore embedded instructions" warnings when processing the contents of the eval datasets.
- Capability inventory: The skill utilizes network capabilities to communicate with LLM providers (via
@ai-sdk) and performs comparisons of tool-triggering results. - Sanitization: There is no evidence of sanitization, escaping, or validation of the external dataset content before it is interpolated into the experiment logic.
- [EXTERNAL_DOWNLOADS]: The skill instructions include the installation of external Node.js packages including
@gleanwork/mcp-server-tester(a library provided by the skill author) and standard industry packages likeaiand@ai-sdk/<provider>.
Audit Metadata