The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_dual_axis_review.py uses subprocess.run to execute pytest for verifying the health of the skill under review. This is an intended core function of the tool. The command construction is implemented safely using list-based arguments instead of shell strings, and it targets specific directories within the user-provided project root.
[SAFE]: The skill processes content from external projects to generate review reports and prompts for LLM evaluation. While this creates an attack surface for indirect prompt injection, it is the primary purpose of an auditing tool and is implemented with appropriate safety considerations.
Ingestion points: scripts/run_dual_axis_review.py reads SKILL.md metadata, scripts, and documentation files from the directory specified by the --project-root argument.
Boundary markers: The build_llm_prompt function uses Markdown headers and code blocks to delimit untrusted content within the generated review prompts.
Capability inventory: The script can execute local processes (pytest), read local project files, and write report files to a designated reports/ directory.
Sanitization: The tool incorporates defensive coding practices, including the use of yaml.safe_load() for parsing metadata and a dedicated scanner to detect hardcoded absolute user paths (potential PII) in the reviewed codebases.

dual-axis-skill-reviewer