NVIDIA / NVIDIA/NemoClaw

[DGX Station][CLI&UX] nemohermes exec returns exit code 0 after an iteration-limit failure

Open
#7,104 1 comment 0 reactions 0 assignees View on GitHub
NV QA
Dominant language
TypeScript
Stars
22.5k
Forks
3.1k
Avg merge
1d 1h
Merged PRs (30d)
715

Description

## NVBug 6471146

**Description**

When Hermes reaches the configured maximum iteration count before completing a tool-driven request, `nemohermes exec` returns exit code 0 even though the requested operation and output artifact are incomplete.

The post-limit summary can also contain a new tool call that is never executed because the agent loop has already stopped. This makes the successful process status inconsistent with the actual task result.

**Steps to Reproduce**

1. Create a Hermes sandbox and configure an MCP server.
2. Submit a tool-driven request that cannot complete within the configured iteration limit:
```
nemohermes {sandbox} exec --timeout 1200 -- \
hermes chat -Q --max-turns 15 -q \
"Use the configured Blender MCP server to complete an OVRTX render and verify the output artifact."
```
3. Allow the request to reach the maximum iteration count.
4. Record the final summary and shell exit code.

**Expected Result**

- Non-zero exit code or explicit incomplete status
- Summary should not contain unexecuted tool calls
- Automation should distinguish completed vs incomplete runs

**Actual Result**

`nemohermes exec` returns exit code 0. The summary contains a proposed tool call that was never executed. The requested artifact is absent.

**Environment**

- DGX Station, Ubuntu 24.04 aarch64
- NemoClaw 0.0.84, Hermes Agent 0.18.0
- Model: Nemotron 3 Ultra 550B

Contributor guide

Open the contributing guide

Research direction

Start by reproducing the iteration-limit case with `nemohermes exec` and the shown `hermes chat` command, recording the shell status and final summary. Trace the CLI entry point that handles the stopped agent loop and summary, then verify that incomplete runs return a non-zero or explicit incomplete status and do not report an unexecuted tool call.

Written by the indexing model from the issue text.

Assessment

Tech stack
typescript
Domain
cli
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
52/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.