The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes extensive project data, including UI components, API routes, and configuration files, to generate audit reports and apply fixes. This creates a significant surface for indirect prompt injection if the files being analyzed contain malicious instructions.
Ingestion points: Reads local files including messages/en.json, Hero.tsx, Features.tsx, Pricing.tsx, auth.ts, and various API route.ts files.
Boundary markers: The instructions lack explicit delimiters or safety instructions to treat project file content as untrusted data rather than behavioral directives.
Capability inventory: The skill has the capability to read files (cat), list directories (find, ls), and most importantly, perform automated edits and write operations across the entire project codebase to "fix" identified issues.
Sanitization: There is no evidence of sanitization or filtering of the file content before it is processed by the LLM logic.
[COMMAND_EXECUTION]: The skill utilizes local shell commands to map out the project structure and perform post-fix validation.
Evidence: Employs find, ls, and cat for discovery, and npx tsc --noEmit for type-checking the project after applying fixes. These commands are executed within a controlled context using hardcoded paths or standard project structures.

product-review