The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied project goals and fetches data from external technology registries (npm, PyPI, crates.io) to formulate a project design and scaffold code. This data flow creates a surface for indirect prompt injection where malicious instructions could potentially be embedded in external metadata or user goals to influence the generated scaffold.\n
Ingestion points: User-provided 'goal' in the /woostack-bootstrap command and live registry metadata fetched via tools like npm view.\n
Boundary markers: An explicit <HARD-GATE> requires manual user approval of the complete design before any filesystem writes or artifact creation occurs.\n
Capability inventory: The skill performs filesystem operations (mkdir, write), network requests (registry lookups, GitHub API), and executes subprocesses (package managers, build tools, test runners).\n
Sanitization: Includes stack validation against requirements and industry research to verify technology choices before presentation.\n- [EXTERNAL_DOWNLOADS]: The skill performs live research and dependency resolution by querying authoritative registries such as npm (using npm view), PyPI, and crates.io. It also interacts with GitHub for optional project publication.\n- [COMMAND_EXECUTION]: To verify the generated project, the skill executes lifecycle commands including dependency installation, linting, formatting, type checking, building, and 'smoke' tests of the application surfaces.\n- [DYNAMIC_EXECUTION]: The skill generates (scaffolds) project scripts and then executes them to verify the build. This involves executing code that was dynamically assembled during the bootstrap process.

woostack-bootstrap