The Agent Skills Directory

[PROMPT_INJECTION]: No malicious prompt injection or behavior override patterns were detected. The skill's instructions establish a clear persona ('react-guardian') and provide architectural directives without attempting to bypass safety protocols or system constraints.
[DATA_EXFILTRATION]: No evidence of unauthorized data access or exfiltration was found. While the skill reads project configuration (e.g., package.json) and analyzes the codebase, these actions are essential to its function as an architectural auditor. No hardcoded secrets or suspicious external network calls are present.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were identified. The included scripts (e.g., 'scan-anti-patterns.ts') are local TypeScript files intended for static analysis of the host codebase. They rely on standard Node.js APIs and reputable dependencies like 'ts-morph'.
[COMMAND_EXECUTION]: The skill contains scripts meant to be executed locally by the agent or user for auditing purposes. These scripts perform safe filesystem operations (reading files, checking sizes) and do not execute arbitrary shell commands or untrusted external code.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it ingests and analyzes untrusted user code. However, the risk is mitigated by the fact that the analysis is performed via static scripts and guided by internal markdown documentation rather than raw execution of user-supplied logic. The severity is assessed as LOW per standard guidelines.
[SAFE]: All external URLs and package references target official documentation, well-known technology vendors (e.g., Vercel, Meta, Microsoft), or reputable open-source repositories. The skill correctly implements security boundaries by surfacing concerns to a specialized 'security-guardian' rather than attempting to perform audits itself.

react-weapon