The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the terminal to execute a Python script (build_report.py) which processes local log files and generates the report.
[DATA_EXPOSURE]: The skill accesses sensitive local databases and configuration files containing interaction history for various agents (Claude, Codex, Gemini, Kimi, and Hermes). This is the intended primary function of the tool.
[SAFE]: The implementation follows security best practices for handling sensitive data, including using read-only mode for SQLite databases and sanitizing all data with HTML escaping to prevent injection vulnerabilities in the generated report.
[SAFE]: The report generator explicitly excludes raw prompt bodies, transcripts, and tool outputs from the final HTML and JSON files to prevent data leakage.
[SAFE]: The skill contains a built-in 'sensitive-pattern scan' that checks all generated content for leaked credentials (API keys, private keys, database URLs) before saving.
[SAFE]: Output files and directories are created with restricted file permissions (0o600/0o700), ensuring only the local system owner can access the generated insights.
[SAFE]: No network connectivity is requested or used; all data collection and processing are performed entirely on the local filesystem.

agent-insights