The Agent Skills Directory

[SAFE]: The skill's primary function is to collect and analyze internal agent metrics to generate performance reports. It does not perform network requests, access sensitive system credentials, or execute external code.
[INDIRECT_PROMPT_INJECTION]: The skill contains a data ingestion surface as it processes metadata generated by other agent tasks. 1. Ingestion points: Reads task summary, deviations, and testResult fields via the TaskGet tool and project plan files via the Read tool. 2. Boundary markers: None identified; the skill assumes task metadata adheres to expected scoring formats. 3. Capability inventory: The skill uses TaskList, TaskGet, and Read tools. It also writes evaluation reports to the .claude/context/reports/ directory. 4. Sanitization: The skill uses logic-based parsing (e.g., extracting pass/fail counts from strings) rather than executing data content, which limits the risk of instruction obedience from malicious metadata.

pipeline-evaluator