The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check.py utilizes subprocess.run to call internal validation utilities and the lifecycle report script. These executions are performed using absolute paths derived from the script's location or specific relative paths within the skill hierarchy. The implementation avoids shell execution and does not process external or unsanitized user inputs, mitigating common command injection risks.
[DATA_EXPOSURE]: The utility scripts/lifecycle_report.py reads SKILL.md files within the repository to evaluate frontmatter and body text for lifecycle signals (e.g., 'draft', 'active'). Although it accepts a skill name as an argument, it restricts file access by specifically looking for a hardcoded filename (SKILL.md) and uses yaml.safe_load for secure parsing of frontmatter.
[PROMPT_INJECTION]: The instructions in SKILL.md and the assertions in evals/evals.json reinforce safety by explicitly prohibiting the agent from automatically promoting skills, editing metadata, or performing mutations. These constraints are designed to prevent the agent from taking autonomous actions based on the reporting output.

skill-lifecycle-manager