The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines an ingestion surface for untrusted data in Phase 0, where it reads user-provided specifications to drive autonomous development. While there are no explicit boundary markers for subagents, the skill implements a 'Review Loop' (Phase 6) and 'Adversarial Verification' steps to check for functional and security issues in the generated code.
[COMMAND_EXECUTION]: The skill executes multiple shell commands for project management, including git init, mkdir, and find. It also runs dynamically resolved test and linting tools such as pytest, ruff, and eslint based on the project's configuration files.
[EXTERNAL_DOWNLOADS]: In Phase 3, the skill may download and install development dependencies like pytest or vitest if they are not found in the environment.
[DATA_EXFILTRATION]: The skill facilitates deployment to external services in Phase 7, including HuggingFace Spaces, PyPI, and GitHub Pages. These network operations are part of the core functionality and target well-known, reputable services.

full-project