The Agent Skills Directory

[COMMAND_EXECUTION]: The main script scripts/run_dual_axis_review.py uses subprocess.run to execute pytest within the directory specified by the --project-root argument. This is the intended functionality for verifying the test health of reviewed skills. The command is constructed as a list, which prevents shell injection.
[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by generating LLM prompts based on the metadata and findings of other skills found in the project root. It uses markdown headings as boundary markers.
[PROMPT_INJECTION]: Mandatory evidence chain for indirect injection: (1) Ingestion points: scripts/run_dual_axis_review.py reads SKILL.md and script files from target projects. (2) Boundary markers: build_llm_prompt uses markdown headers (e.g., ## Skill, ## Files to inspect) to separate external content. (3) Capability inventory: The skill can execute pytest and read/write local files. (4) Sanitization: External content is interpolated without explicit sanitization.
[COMMAND_EXECUTION]: The test file scripts/tests/conftest.py uses importlib.util to dynamically load the main script for unit testing purposes, which is a standard development practice.

dual-axis-skill-reviewer