The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The board simulation processes user-supplied proposals and context by interpolating them into individual director prompts without proper boundary markers.
Ingestion points: The proposal and context parameters are injected into director-specific Task prompts in SKILL.md.
Boundary markers: The prompt templates do not use clear delimiters or "ignore instructions" warnings to separate the untrusted proposal content from the agent's specialized role definitions.
Capability inventory: The skill has access to run_shell_command for directory management and write_file for persisting session data.
Sanitization: No input sanitization or validation is applied to the user-provided proposal before it is processed by the simulation.
[COMMAND_EXECUTION]: The skill leverages shell commands to initialize its internal file-based message bus and persistence layer.
Evidence: SKILL.md instructs the agent to use run_shell_command to execute mkdir -p for creating session directories in the conductor/tracks/ path.

board-of-directors