The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided ideas and reads local documentation (e.g., plan-critic.md, interview-engine.md) to facilitate dialogue. While this ingestion represents a theoretical attack surface for indirect prompt injection, the risk is mitigated by the skill's restricted environment.
Ingestion points: User input in chat and file system access via Read, Glob, and Grep tools.
Boundary markers: No specific boundary markers or delimiters are defined for processing external data.
Capability inventory: The skill is limited to Read, Glob, Grep, Agent, and AskUserQuestion. It explicitly omits Write, Edit, and Bash to prevent file modifications or command execution.
Sanitization: No explicit sanitization is described, but the architectural restriction to read-only tools and a read-only subagent ('Explore') significantly limits the impact of any potential injection.
[COMMAND_EXECUTION]: The skill's configuration explicitly omits the Bash tool and restricted subagent types to ensure no shell commands can be executed during the brainstorming process.

brainstorm