-
Notifications
You must be signed in to change notification settings - Fork 0
Troubleshooting
Your shell has not picked up the docker group. Run newgrp docker, open a fresh shell, or
prefix commands with sudo.
The target URL is wrong. For a custom HTTP agent, point --target at the chat endpoint
(e.g. http://host/chat), not the root. See Targets.
The model/provider isn't reachable or served:
- Pass a model your backend actually serves (
--model), and--api-basefor self-hosted servers. See LLM Providers. - Large models on a slow server can exceed the request timeout. Use a smaller/faster model,
or
--mode seed(no LLM).
Reinstall — the data ships inside the package (aphasia/data/). uvx --refresh aphasia-agentry
or pip install -U aphasia-agentry.
A weak LLM phrases fewer triggering payloads. Use --mode hybrid (seeds first) or a stronger
model. --mode seed proves all 11 against the fixture deterministically. See Attacker Modes.
A target/provider error occurred and nothing was proven. Check the printed message and
<run-dir>/evidence.jsonl for the failing scenario(s). See CLI Reference for exit codes.