Deep Agent Review
How the LangChain Deep Agent fits into the deterministic scanner path.
What this is
Deep Agent review explains scanner artifacts and checks whether claims are grounded in local evidence.
When to use it
Use Deep Agent review after a scan has already produced deterministic artifacts.
Command or example
export OPENROUTER_API_KEY=<key>
uv run --extra deep-agent agent-permit investigate .agent-permit/runs/<run_id>Output to expect
The investigation writes a cited Markdown report and usage metrics. The citation critic checks whether claims are supported by local artifacts.
How to interpret it
Deep Agent is not the scanner. It is the reviewer assistant. It reads bounded artifacts, explains why a finding matters, and points back to evidence.
Common mistakes
- Letting model output create findings without scanner evidence.
- Treating uncited model claims as approved facts.
- Running live model review without cost and cache telemetry.