The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes an initialization script (scripts/idea-refine.sh) that sets up a local directory for storing ideas. This script is benign, contains no external network calls, and does not process untrusted input.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided ideas and scans the local codebase using Glob, Grep, and Read tools. This creates an attack surface where malicious content in the codebase could theoretically influence the agent's output. However, the risk is minimal as the skill's primary actions are conversational and file writes are restricted to a dedicated path (docs/ideas/) requiring user approval.
Ingestion points: User input ($ARGUMENTS) and project files (via codebase scanning tools).
Boundary markers: The skill relies on natural language instructions to separate phases but does not use explicit technical delimiters for external data.
Capability inventory: The skill uses tools for reading files and directory structures, and performs file writing to the local filesystem.
Sanitization: No explicit sanitization of codebase content is mentioned before it is processed by the LLM.

idea-refine