The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs agents to treat accepted decision records as binding instructions for work within the repository (documented in templates/AGENTS.md). This creates an indirect prompt injection surface because these records originate from user-contributed drafts that are not sanitized for instructional content.
Ingestion points: The agent processes content from docs/decisions/drafts/ and accepted record directories.
Boundary markers: Absent. The agent is directed to follow the records as authoritative rules without specific delimiters to isolate potential adversarial instructions.
Capability inventory: The bundled decisions.py script performs file system operations including writing, deleting, and symlinking files.
Sanitization: The decisions.py tool validates metadata format but does not sanitize the markdown body or rationale fields for potential prompt injection patterns.
[COMMAND_EXECUTION]: The install command in scripts/decisions.py automatically modifies the repository's .git/hooks/pre-commit file and changes its permissions (chmod 0o755) to execute the script on every commit attempt. It also creates a machine-specific symlink for the tool within the scripts/ directory.

decision-records