The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to process data from hosted evaluation runs, which acts as an ingestion point for untrusted external data. Malicious evaluation cases could theoretically contain instructions designed to influence the agent during the triage process.
Ingestion points: Data is ingested through tools like get_eval_run_steps, list_eval_run_iterations, and list_eval_cases (specifically prompts, scenarios, and expectedOutput fields).
Boundary markers: The skill does not define specific delimiters or instructions for the agent to ignore embedded commands within the evaluation data.
Capability inventory: The agent has the capability to perform network operations (via the hosted eval tools) and follow external evidence URLs (screenshots/videos) provided in the test results.
Sanitization: No mention is made of sanitizing or escaping the content of the evaluation steps before analysis.

run-mcpjam-evals