The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands like git, ls, grep, mkdir, and find for managing project files and branch state throughout the development lifecycle.
[EXTERNAL_DOWNLOADS]: The research phase involves fetching information from web sources and official documentation. The instructions explicitly mandate treating this content as untrusted input and stripping out any directives or instructions it might contain.
[REMOTE_CODE_EXECUTION]: The workflow involves executing the project's own development scripts (e.g., npm test, npm run lint) to verify implementation tasks. These commands are dynamically identified from the project's manifest files.
[PROMPT_INJECTION]: The skill processes user-provided PRDs and external research content, creating an indirect prompt injection surface. To mitigate this, the instructions require the agent to transform and summarize findings in its own words rather than using verbatim text.

spec-driven