The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied research questions and external project files which are interpolated into prompts for multiple sub-agents.
Ingestion points: Input originates from the $ARGUMENTS variable and project files retrieved via the Read tool.
Boundary markers: The skill uses specific template tags (e.g., [QUESTION], [LABEL], [DISCIPLINE]) to delimit research data from agent instructions.
Capability inventory: The skill uses standard file system tools (Read, Write, Edit, Glob, Grep) and agent orchestration (Task) within the project environment.
Sanitization: Content is processed as-is for analysis purposes, with safety provided by the platform's underlying model guardrails and the specific persona-driven constraints applied to the agents.
[COMMAND_EXECUTION]: The skill mentions an optional 'Council Mode' that interfaces with a council-cli for multi-model synthesis. This utility is referenced as part of the shared workspace protocol for high-diversity research synthesis.

multi-perspective