The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data by reading arbitrary files from the user's project directory as part of its multi-phase audit process.
Ingestion points: Reads project files including routing configurations, component source code (React, Vue, React Native), and environment files (.env) across all phases.
Boundary markers: The instructions lack explicit delimiters or instructions to ignore embedded prompts within the analyzed codebase, which could allow malicious code comments or strings to influence the agent's behavior.
Capability inventory: The skill has extensive file-read and file-write capabilities (documented in Phase 8 and the Fix Protocol). It does not request network access or the ability to execute the code it audits.
Sanitization: No validation or sanitization is performed on the content read from the codebase before it is used to generate reports or suggested fixes.
[COMMAND_EXECUTION]: The skill is designed to modify the local filesystem through its 'Fix Protocol,' which writes code changes back to the user's files based on identified bugs. While this is an intended feature, it represents a high-privilege capability that operates on data derived from untrusted source files.

beta-tester