The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from requirement specifications and exported review logs, but implements layered mitigations.
Ingestion points: Requirements markdown files are read by scripts/init-mockup.mjs, and review logs are loaded into the browser-based prototype engine from comments.js.
Boundary markers: references/agentic-setup.md and references/agentic-setup.md contain explicit instructions for the agent to treat input as data and ignore any embedded commands or instructions attempting to change scope.
Capability inventory: The skill has the ability to write to the filesystem to create prototype scaffolds and execute git via scripts/sync-tokens.mjs to resolve the repository root.
Sanitization: The skill implements robust path validation in scripts/sync-tokens.mjs and scripts/init-mockup.mjs using regular expressions and containment checks to prevent directory traversal and access to sensitive folders like .git or .uxproof.
[SAFE]: No malicious patterns were detected. The skill demonstrates high security hygiene by:
Hardcoding safe arguments for subprocess calls (e.g., git rev-parse).
Storing review data locally in the browser's localStorage rather than exfiltrating it.
Providing clear documentation on secret hygiene and avoiding the use of production credentials.

om-ux-design