The Agent Skills Directory

[COMMAND_EXECUTION]: The skill references a local helper script scripts/idea-refine.sh to initialize the docs/ideas/ directory. The script uses standard commands (mkdir) and does not perform any dangerous operations or network requests.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and analyze untrusted data provided by the user and discovered within the codebase. * Ingestion points: User-provided ideas via $ARGUMENTS and local files accessed via Glob, Grep, and Read tools (SKILL.md). * Boundary markers: None explicitly defined in the instructions to isolate ingested content from logic. * Capability inventory: Shell command execution (bash), file creation (docs/ideas/*.md), and read-access to the codebase (Glob, Grep, Read). * Sanitization: No specific input sanitization or output encoding is described.
[SAFE]: All external URLs point to a reputable public repository on GitHub. No malicious patterns, obfuscation, or unauthorized data exfiltration were detected.

idea-refine