The Agent Skills Directory

[SAFE]: The skill implements a coordinated multi-agent workflow. It limits the execution environment by explicitly listing permitted sub-agents in the agents frontmatter field, adhering to the principle of least privilege.
[DATA_EXPOSURE]: The instructions direct the agent to read local project documentation (AGENTS.md, CLAUDE.md) and source code to gather context for the brainstorming session. This is a standard operational requirement for development-focused skills and does not involve unauthorized data access or external exfiltration.
[COMMAND_EXECUTION]: The skill uses a simulated command-line interface within its markdown instructions (e.g., --profile, --seats). These parameters are interpreted by the agent to determine logic flow and agent selection rather than being executed in a system shell, which avoids command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external configuration files (commands/brainstorm-menu.json) and project source code. While this represents a potential surface for indirect prompt injection, the impact is mitigated by the skill's restricted sub-agent list and its primary role as a collaborative design tool.
Ingestion points: commands/brainstorm-menu.json, project source code files.
Boundary markers: None explicitly defined for external content ingestion.
Capability inventory: File writing to docs/features/ and invocation of specified scout agents via the Agent tool.
Sanitization: No explicit sanitization logic for ingested JSON or code content.

define-cognitive-brainstorm