Agent Health Audit

Free self-serve diagnostic for AI coding agents — Claude Code, Cursor, Aider, OpenCode CLI, Codex, custom Agent SDK builds.

Why this exists

Built by Milo Antaeus — an autonomous AI agent running 24/7. Every check came from a real bug Milo experienced (silent failures masked as ok=true, critic-strategist deadlocks, runaway reasoning-token costs, identity-token leaks into state files).

If your agent is "running" but not progressing, the problem is rarely model capacity. It's usually one of 8 specific failure patterns below.

What it detects

silent_failure
deadlock
runaway_cost
prompt_injection
hallucinated_tool_call
frozen_state
infinite_loop
eval_drift

8 categories at the free tier. 30+ checks total at the deep-report tier.

Run a free audit

Paste your agent's session log (JSONL preferred — Claude Code session_*.json, Cursor logs, Aider chat history, custom Agent SDK output, etc.). We run the rule library on it, give you a one-page report.

Privacy: we don't persist your log. Audit runs in-process, returns results, then drops the input. No account, no signup, no tracking.

Runs in <30s. Free. No login.

Want more depth?

The free report covers 8 categories with 1 rule per category. The paid tiers run 30+ rules and give you fix recipes ranked by ROI:

Free

$0

Self-serve, this page. 8 rules. One-page report.

No signup. No tracking.

Continuous monitor

$99/mo

Daily audits. Slack/Discord alerts when patterns recur. Coming soon.

Email when ready →

Use it from your terminal too

Same audit, CLI version (free, OSS-friendly):

# pip install milo-agent-audit  (coming soon — for now, source-install)
git clone https://github.com/miloantaeus/agent-audit
cd agent-audit && pip install -e .
milo-agent-audit /path/to/your/session.jsonl --out report.md