The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and follow the structure of existing "peer skills" in the repository to author new ones. This creates a surface where a malicious peer skill could influence the structure or content of newly generated skills.
Ingestion points: Peer SKILL.md files located within skills/<category>/ (Workflow step 1).
Boundary markers: Standard YAML frontmatter delimiters (---) are specified as hard requirements.
Capability inventory: The skill utilizes write_file, patch, and skill_manage for file operations, along with git for version control.
Sanitization: The provided validation workflow explicitly utilizes yaml.safe_load for parsing, which prevents unsafe YAML deserialization during the validation process.
[DYNAMIC_EXECUTION]: The skill provides a Python snippet intended for local validation of SKILL.md files.
Evidence: Step 4 of the Workflow section provides an example script that uses import yaml, re, and pathlib to read and validate the frontmatter.
Context: The execution is restricted to standard validation checks and uses yaml.safe_load to ensure that data is loaded without executing arbitrary code.

hermes-agent-skill-authoring