The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes the $ARGUMENTS placeholder to interpolate user input into the context provided to the primary agent and its sub-agents. This allows a user to potentially provide instructions that attempt to override the skill's intended behavior or influence the reasoning of the committee agents.
[INDIRECT_PROMPT_INJECTION]: The skill functions as a conduit for untrusted user input, passing it to sub-agents created through the create_agent tool.
Ingestion points: User input is captured via the $ARGUMENTS variable in SKILL.md and included in the problem-level prompt.
Boundary markers: The skill includes a mandatory 'no-edits' suffix ('This is analysis only. Do NOT edit, create, or delete any files. Do NOT write code.') that is appended to every prompt sent to committee members, providing a defensive boundary against instruction-based attacks on the sub-agents.
Capability inventory: The skill utilizes the list_profiles tool to read agent configurations and the create_agent tool to execute secondary reasoning processes.
Sanitization: No explicit validation, filtering, or escaping of the $ARGUMENTS string is performed before it is interpolated into the prompts.

paseo-committee