The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides Node.js scripts (audit-rolecard-security.mjs and check-security-baseline.mjs) intended for local execution to perform static code analysis. These scripts operate on a user-provided target directory to generate security reports.
[INDIRECT_PROMPT_INJECTION]: The skill identifies an attack surface where it processes untrusted external data (SillyTavern rolecards) which may contain malicious instructions.
Ingestion points: Files (JSON, JS, HTML, etc.) are read from a target directory using fs.readFileSync in scripts/audit-rolecard-security.mjs.
Boundary markers: The instructions explicitly state that rolecard content is treated as an executable supply-chain surface and should be scanned read-only without evaluation. The scanner uses a custom JSON walker to ensure strings are not interpreted as code.
Capability inventory: The scripts utilize file system read/write access (for reports) and directory traversal. There is no evidence of network operations or shell execution of the data being scanned.
Sanitization: The auditing script specifically redacts potential credential values and parser error excerpts in its output to prevent the exposure of sensitive data in the audit report.
[SAFE]: The skill demonstrates security best practices by including a SENSITIVE_NAMES filter to skip system-level sensitive files (like .env or SSH keys) during scans and provides a clear threat model for its operations.

sillytavern-rolecard-security