The Agent Skills Directory

[SAFE]: The skill serves a legitimate developer productivity and security auditing purpose. It provides comprehensive rules for reviewing TanStack Start applications, covering client/server boundaries, server functions, and authentication. All referenced documentation links point to official, well-known domains (tanstack.com, typescriptlang.org, zod.dev).
[NO_CODE]: The skill consists entirely of Markdown instructions and rule definitions. It contains no executable scripts, binaries, or automated installation steps.
[PROMPT_INJECTION]: While the skill uses an 'adversarial' persona for its subagent reviewer, this is explicitly scoped to the code review task ('find violations... not to confirm compliance'). There are no attempts to bypass system safety filters or extract system prompts.
[DATA_EXPOSURE]: The skill is designed to read project configuration (tsconfig.json, router.tsx) and code diffs to perform its function. This access is necessary for the stated purpose and does not involve exfiltration of sensitive files like SSH keys or AWS credentials.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of code diffs or file contents provided for review. 1. Ingestion points: The {{TARGET_CONTENT_OR_PATHS}} placeholder in references/reviewer-prompt.md is the entry point for external data. 2. Boundary markers: The prompt template uses Markdown headers and distinct sections to separate the target content from the rules and instructions. 3. Capability inventory: The skill uses file reading capabilities and subagent task invocation to render a verdict. 4. Sanitization: There is no explicit sanitization of the input code, which is standard for static analysis tools; however, the structured prompt helps constrain the subagent's behavior.

adversarial-tanstack