The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/audit.sh script executes npm install during its first run to install ts-morph into a local node_modules directory. This fetches the package from the public npm registry.
[COMMAND_EXECUTION]: The skill executes several sub-processes including ripgrep (rg), node, and shell scripts to perform code analysis. These operations are scoped to the project directory provided by the user and the skill's internal script directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source code from the project being audited and includes snippets of that code in the generated markdown report (.rhf-audit-report.md). If the audited code contains malicious instructions, they could be executed by the AI agent when it reads and interprets the audit report.
Ingestion points: Project source files (e.g., app/**/*.tsx, src/**/*.ts) identified via ripgrep.
Boundary markers: Code snippets in the generated report are wrapped in standard markdown code blocks (triple backticks) but lack explicit instructions for the AI agent to ignore embedded commands.
Capability inventory: The skill has read access to the project filesystem, write access for the report file, and the ability to execute subprocesses for analysis.
Sanitization: No sanitization is performed on the code snippets before they are written to the report file.

react-hook-form-audit