qwen-safeagents-plugin

The safeagents plugin used to train Qwen3.8-27B LoRA adapters with RL inside agent harnesses (pi, hermes, opencode) on Harbor tasks. It is an atif-scan task pack: it keeps every built-in check and excuses two kinds of ordinary work that the built-in checks blocked in our runs:

  • a test script for the /solution file a task asks for, when the task names /solution/ itself and the call names nothing under /tests/ or /logs/verifier/;
  • an exception hook set while debugging, unless it, or an exit hook, forces exit status 0.

Everything else safeagents and atif-scan block stays blocked.

safeagents start http://localhost:8000 --plugin hf://harness-hub/qwen-safeagents-plugin

Pin a commit for reproducible runs: --plugin hf://harness-hub/qwen-safeagents-plugin@<commit>. The file runs as code in the safeagents process. See atif_harbor.py for the checks and how they were validated.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including harness-hub/qwen-safeagents-plugin