The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash with a restricted set of allowed commands (ls*, mkdir*, date*) to manage the logging directory and entry dates, which is appropriate for its stated purpose.
[DATA_EXPOSURE]: The skill reads and writes to local files (log/ideas.md, log/.ideas-state.json). It does not perform any network operations or access sensitive system directories.
[INDIRECT_PROMPT_INJECTION]: The skill processes 'raw ideas' which are untrusted data.
Ingestion points: Untrusted text enters through the capture argument or interactive input.
Boundary markers: The skill uses a structured Markdown template and specific status tags to delimit entries.
Capability inventory: Capabilities include writing to the local log and making approved file edits during the integrate phase.
Sanitization: The skill relies on user approval via AskUserQuestion before acting on any 'Direct proposal' derived from an idea, providing a human-in-the-loop safety check.

ideas