The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool across various roles (profiler, benchmarker, optimizer) to execute standard developer utilities. This includes running language-specific benchmark commands such as cargo bench, go test -bench, and make bench, as well as performing build analysis and performance profiling. These operations are consistent with the skill's stated purpose of performance optimization.
[PROMPT_INJECTION]: The skill ingests local codebase files as part of its analysis phase. While processing untrusted data (source code) can pose an indirect prompt injection risk, the skill includes a multi-stage review process (via the 'reviewer' role) that acts as a mitigation by validating generated code changes before they are finalized.
Ingestion points: Source code files are read by the profiler and optimizer roles (roles/profiler/role.md, roles/optimizer/role.md).
Boundary markers: The skill does not explicitly use boundary delimiters for the ingested code content in its prompts.
Capability inventory: The system has broad capabilities, including the ability to write and edit files (optimizer role) and execute shell commands (profiler and benchmarker roles).
Sanitization: No automated sanitization of ingested code content was identified; however, the structured multi-agent review workflow provides a procedural layer of safety.

team-perf-opt