agent-ledger: a security and audit plugin for Hermes Agent

Описание и применение

agent-ledger — локальный плагин для Hermes Agent, записывающий все вызовы инструментов в SQLite‑базу, проверяющий их выполнение, выявляющий «призрачные» вызовы и оценивающий риск поведения агента по OWASP Agentic Top 10, выдавая 0–100 балл и список обнаруженных угроз.

Задача

When running an AI agent locally, users trust it to edit files, restart services, and push to GitHub, but lack visibility into what the agent actually did and how risky its behavior is.

Результат

The plugin provides a local, telemetry‑free audit trail of tool calls, detects ghost calls, and assigns a reproducible posture score based on OWASP criteria, enabling users to understand and mitigate agent risk.

Рабочий процесс

• Clone the agent-ledger plugin into ~/.hermes/plugins/agent-ledger
• Restart the Hermes gateway to start recording tool calls
• Run python3 ~/.hermes/plugins/agent-ledger/posture.py to generate a posture report
• Optionally set up posture_daily.py with a Telegram bot token and cron for daily digests

Доказательства

Every tool call your agent makes, recorded locally before it runs — with a 0–100 posture score and an honest list of what it can't see.

Источник

When a ghost is detected, the plugin injects a warning into the agent's next context, so the agent itself goes back and verifies reality instead of trusting its memory.

Источник

score = 100 − Σ{CRITICAL:25, HIGH:15, MEDIUM:8, LOW:3} over all findings, no deduplication, and the full finding list always comes back so you can recompute it by hand.

Источник

The scorer ships with its full test suite, and you're encouraged to run it before trusting it: ... PASS

Источник