The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): The instructions use directive language ("You MUST") to ensure the agent follows the brainstorming process, but do not attempt to bypass core safety filters or ignore system instructions.
[DATA_EXFILTRATION] (SAFE): The skill reads local project files and commit history to gain context, which is necessary for its stated purpose. It writes documentation to local paths within the project. No unauthorized network transmission or credential harvesting was detected.
[REMOTE_CODE_EXECUTION] (SAFE): No remote scripts or external package installations are defined. The skill references other internal agent skills for specific tasks (e.g., git worktrees) rather than executing arbitrary remote code.
[COMMAND_EXECUTION] (SAFE): The skill triggers standard git operations (commit) and internal tool calls (AskQuestion), which are consistent with its documented purpose and occur within the local project context.
[DATA_INGESTION_SURFACE] (LOW):
Ingestion points: Reads project files, documentation, and recent commit messages (SKILL.md).
Boundary markers: Absent; there are no specific instructions to ignore embedded prompts in the project data.
Capability inventory: Can write design files to docs/plans/ and perform git commits.
Sanitization: Absent; the skill relies on the LLM's inherent processing of the project context to generate documentation.
Risk: While an indirect injection could occur via malicious project files, the impact is limited to the generation of design documents and does not enable system-level compromise.

brainstorming