Harness System

[hunglv53]

I wanna build the Coding Agent Harness. Does the Agno support that ?

The current agent from Agno, it seems to be that does not run the long running tasks, get the feedback, auto correct when having the error.

thank you

[barry166]

Agno can be one layer inside a coding-agent harness, but it is not the whole harness by itself.

For a coding harness you usually need an outer control loop around the agent: create task -> run agent step/tool call -> execute shell/tests -> capture errors -> feed the result back -> stop on success/budget/approval. Agno can model the agent, tools, memory, and team pieces, but the long-running retry loop, workspace isolation, test runner, and approval policy are things you should build around it.

So the shape would be: Agno agent/team for planning and tool selection, plus your own orchestrator for repo checkout, command execution, test feedback, retries, and termination conditions. If you need Codex/OpenHands-style autonomous repair, treat Agno as the agent runtime component, not the full harness product.

[nick211908]

https://github.com/shareAI-lab/learn-claude-code/

There is a lot in harness from 1.memory 2.context-compation 3.hooks 4. skill-loading 5. Subagents you can learn it using the above repo.