The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a governance framework for project rules that modify agent behavior, creating an indirect prompt injection surface.\n
Ingestion points: The skill ingests data from the .agents/rules/ directory, including the index file and individual rule definitions.\n
Boundary markers: Structural Markdown requirements (specific headers and fields) are enforced, though the rule content itself is free-form text.\n
Capability inventory: The skill includes scripts capable of reading/writing files and executing git commands.\n
Sanitization: Metadata like rule IDs and levels are validated, but the normative rule text is not sanitized.\n- [COMMAND_EXECUTION]: The provided get-rules.mjs script executes the system git binary to interact with the repository.\n
Evidence: The script uses execFileSync to run git commands for repository status and file retrieval. It performs path resolution and validates commit identifiers to prevent command injection.\n- [DATA_EXPOSURE]: The skill manages files within the .agents/rules/ project directory.\n
Evidence: The init-rules.mjs and get-rules.mjs scripts perform directory creation and file reading within the project root. Access is restricted to the rule repository structure.

rule-steward