The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data including PRD (Product Requirement Document) files, User Journey documents, and prototype source code. This represents an indirect prompt injection surface.
Ingestion points: PRD files, User Journey files, _prototype-manifest.md, and actual prototype code files scanned via Glob/Grep.
Boundary markers: The skill relies on a manifest file (_prototype-manifest.md) and a defined sandbox directory to scope its analysis.
Capability inventory: The skill uses file-system reading tools (Read, Glob), search tools (Grep), and version control tools (git diff, git status) to perform its review.
Sanitization: The skill's instructions emphasize using the data solely for verification and reporting purposes, with a 'Reviewer only' stance that forbids modifying any code.
[COMMAND_EXECUTION]: The skill instructions include the use of shell-based utilities like git diff, git status, Glob, and Grep. These are used appropriately within the context of a code auditing tool to verify file changes and directory structure consistency.

prototype-reviewer