The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform environment discovery and audit tasks, such as checking file sizes in node_modules and searching for configuration files. This functionality is consistent with its stated purpose of performing a technical audit.
[EXTERNAL_DOWNLOADS]: The skill uses WebFetch and WebSearch to gather information, likely for comparing dependencies against external performance data (e.g., bundle size benchmarks). All observed network operations are aligned with the primary functionality.
[PROMPT_INJECTION]: As an audit tool, the skill ingests untrusted data from the local project, including source code and configuration files (package.json, tsconfig.json). While this presents a surface for indirect prompt injection where maliciously crafted project files could attempt to influence agent output, this is an inherent risk of auditing third-party code.
Ingestion points: Project configuration files and source code retrieved via Read, Glob, and Grep tools.
Boundary markers: None explicitly defined to separate untrusted file content from instructions.
Capability inventory: Access to Bash (command execution), Read (file access), and WebFetch (network requests).
Sanitization: No explicit sanitization or validation of the processed file content is described in the instructions.

prism-audit