The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a provided Node.js script (scripts/redact-preview.mjs) to perform file system analysis. This script is restricted to reading local files using standard Node.js APIs (fs.readFileSync, fs.readdirSync) and does not invoke subprocesses or execute arbitrary shell commands.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it reads and processes user-specified files to detect sensitive patterns.\n
Ingestion points: File content is ingested in scripts/redact-preview.mjs through the file paths provided in the target argument.\n
Boundary markers: The script implements a safety-first output policy by only reporting metadata (line numbers and finding categories) to the agent context, explicitly avoiding the inclusion of the matched sensitive values themselves.\n
Capability inventory: The script is limited to read-only file access (fs.readFileSync) and regex matching. It lacks capabilities for file writing, network communication, or dynamic code execution.\n
Sanitization: Content is treated strictly as raw text for pattern matching; the script does not evaluate or interpret the file content as instructions.\n- [DATA_EXFILTRATION]: Although the skill identifies sensitive data such as secrets and tokens, it does not include any network operation capabilities or external transmission logic. It is designed as a defensive tool to help users keep sensitive data within their local environment.

llm-wiki-privacy-redactor