The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute local Python scripts and the claude CLI for evaluation tasks.
Evidence: SKILL.md references scripts such as run_eval.py, run_loop.py, and quick_validate.py for performing various evaluation modes.
Evidence: The skill documentation explicitly requires the claude CLI for trigger testing operations.
[PROMPT_INJECTION]: The skill's evaluation agents process transcripts and output files from other skills, which constitutes an indirect prompt injection surface.
Ingestion points: agents/grader.md, agents/comparator.md, and agents/analyzer.md read execution transcripts and output directories containing data from potentially untrusted skills.
Boundary markers: The instructions for these subagents do not specify any delimiters or warnings to ignore malicious instructions embedded within the data being graded.
Capability inventory: The skill possesses Bash, Write, and Agent tools, which could potentially be abused if a subagent is compromised by malicious input.
Sanitization: There is no evidence of content sanitization or validation for the transcripts and outputs before they are processed by the grader or analyzer subagents.

skill-eval