The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation specifies the integration of shell-based hooks that execute local Python scripts, such as python3 ~/.claude/hooks/council_hook.py, as a core part of its workflow functionality.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by processing untrusted data—specifically operation text and context—within a deliberation process that governs high-risk actions like production deployments and API calls.
Ingestion points: Untrusted data enters the system through the operation_text and context parameters in the detect_trigger and convene_council functions defined in SKILL.md.
Boundary markers: There is no documentation regarding the use of boundary markers or specific instructions for agents to ignore instructions embedded within the metadata.
Capability inventory: The deliberation result influences high-risk "External Commitments" including system deployments, API publishing, and architectural migrations.
Sanitization: The skill description does not indicate input validation, escaping, or sanitization of external content prior to its use in the council deliberation cycle.

dwa-council