The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and validate other agent skills, which involves ingesting content from untrusted files. This creates a potential surface for indirect prompt injection.\n
Ingestion points: The scripts/validate.py script reads and parses SKILL.md, reference files, and bundled scripts from a user-provided directory. The agent is also instructed in SKILL.md to read linked files in full for manual auditing.\n
Boundary markers: Absent; the instructions do not specify the use of delimiters or warnings when the agent reads external file content.\n
Capability inventory: The agent has the ability to read files and execute the local validation script. The script itself performs regex-based checks and YAML parsing but does not execute external code.\n
Sanitization: The validator employs yaml.safe_load() for frontmatter parsing, protecting against YAML-based code execution.\n- [COMMAND_EXECUTION]: The skill requires the agent to execute a bundled Python script to perform its primary validation task.\n
Evidence: SKILL.md instructs the agent to run uv run scripts/validate.py <skill-dir>.\n- [EXTERNAL_DOWNLOADS]: The validation script relies on a common third-party library for parsing.\n
Evidence: scripts/validate.py declares a dependency on pyyaml>=6.0 via PEP 723 inline metadata.

writing-skills