The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to utilize various system utility and profiling commands to investigate performance limiters and verify environment states.
Evidence: Mentions usage of uptime, nproc, top, pidstat, node --cpu-prof, py-spy, perf, and strace -c in the 'Before you run anything' and 'The questions' sections of SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external performance data, PR descriptions, and measurement scripts provided by users, which could potentially contain malicious instructions.
Ingestion points: Performance claims, measurement scripts, and benchmark outputs provided in the session context (SKILL.md).
Boundary markers: None present to distinguish untrusted user data from instructions.
Capability inventory: Suggested execution of shell-based diagnostic and profiling tools (SKILL.md).
Sanitization: No explicit instructions for sanitizing or escaping content from measurement scripts before analysis.

benchmark-checklist