The Agent Skills Directory

Command Execution: The helper script scripts/list_skills.py uses subprocess.run to execute the git command to identify the repository root. This is a standard and legitimate use of shell commands for repository discovery and management.
Indirect Prompt Injection: The skill reads SKILL.md files from the repository to generate an index. Because these files are external to the skill's own code, they represent a potential surface for indirect instructions. 1. Ingestion points: The _rows_from_dir function in scripts/list_skills.py reads data from SKILL.md files located in the repository root and plugin cache. 2. Boundary markers: The skill does not employ specific boundary markers when outputting extracted skill descriptions to the agent. 3. Capability inventory: The skill environment includes basic shell execution for repository discovery but lacks file-writing or external network capabilities. 4. Sanitization: The parser extracts frontmatter fields without specific sanitization or escaping of the description content.
Dynamic Execution: The SKILL.md frontmatter includes a pre-flight block that executes a local Python module (setup_preflight). This is a standard environmental validation step used by the underlying framework.

list-skills