The Agent Skills Directory

[SAFE]: The skill defines a procedural workflow for UI development. Its primary operations involve reading local project metadata, managing Git-tracked documentation files, and using a helper script for environment discovery. All operations are confined to the local project environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local file system (project configuration and parity logs) which could theoretically be used for injection, although the implementation uses safe parsing methods.
Ingestion points: The scripts/prepare-prototype-work.sh script reads package.json files found in the designs/ directory. The workflow also reads .parity/parity-map.md and .parity/ledger.md to track progress.
Boundary markers: Data from package.json is processed via JSON.parse() within a Node.js one-liner, ensuring a strict boundary between data and code. Markdown content is treated as instructional data for the agent.
Capability inventory: The skill can read local files, execute local Bash scripts, and invoke Git commands. It mentions a parity-verifier tool for visual diffing.
Sanitization: The prepare-prototype-work.sh script validates the structure of package.json and checks for specific string keys (react, scripts.dev) before proceeding, effectively filtering out non-compliant or malformed data.

prototype-backed-workflow