The Agent Skills Directory

[SAFE]: The skill is designed for planning and exploration. It explicitly instructs the agent to wait for user confirmation before writing documentation or generating task breakdowns, and it strictly prohibits applying any changes to the project during the brainstorming phase.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a standard attack surface where it reads project context (files, documentation, and git history) which could contain untrusted instructions. However, the design of the skill (one question at a time, incremental validation) minimizes the impact of potential malicious content.
Ingestion points: Project root directory (files, docs, git commits) analyzed via subagents.
Boundary markers: Absent in the instructions.
Capability inventory: File read access, file write access limited to the docs/ directory, and invocation of other agent skills.
Sanitization: Not explicitly performed; the skill relies on user oversight and confirmation steps to mitigate risk.

brainstorming