fix(onboard): propagate context window, max tokens, and reasoning to sandbox image by javlund · Pull Request #1956 · NVIDIA/NemoClaw

javlund · 2026-04-16T07:56:50Z

Summary

The NEMOCLAW_CONTEXT_WINDOW, NEMOCLAW_MAX_TOKENS and NEMOCLAW_REASONING variables were not sent all the way to the openclaw.json file (and thus the agent configuration), I've tried to fix that. As the Dockerfile is altered, this will likely require a sandbox re-creation to kick in.

There were a few tests I couldn't get green as they were timing out, but they don't seem to be related to the files changed in this PR.

Changes

adds 3 ARGs to the Dockerfile with the same defaults as before, promotes them to ENV, and the Python script reads from os.environ instead of literals
patchStagedDockerfile() in onboard.ts reads the 3 env vars from the host and patches the ARG lines before image build
Unit tests to verify

Type of Change

Code change (feature, bug fix, or refactor)
Code change with doc updates
Doc only (prose changes, no code sample modifications)
Doc only (includes code sample changes)

Verification

npx prek run --all-files passes
npm test passes
Tests added or updated for new or changed behavior
No secrets, API keys, or credentials committed
Docs updated for user-facing behavior changes
make docs builds without warnings (doc changes only)
Doc pages follow the style guide (doc changes only)
New doc pages include SPDX header and frontmatter (new pages only)

AI Disclosure

AI-assisted — tool: GitHub Copilot

Signed-off-by: Jacob Avlund jacob@avlund.dk

Summary by CodeRabbit

New Features
- Context window size, maximum token limits, and reasoning mode can now be configured via environment variables for more flexible deployments and builds.
Tests
- Added automated test coverage to ensure environment-based build staging correctly applies those configuration values.

…sandbox image

coderabbitai · 2026-04-16T07:58:03Z

📝 Walkthrough

Walkthrough

Added three Nemoclaw build args/env vars—NEMOCLAW_CONTEXT_WINDOW, NEMOCLAW_MAX_TOKENS, NEMOCLAW_REASONING—and propagated them into the image ENV and generated OpenClaw config. The onboard patcher now validates and injects these values into staged Dockerfiles; a test verifies the behavior.

Changes

Cohort / File(s)	Summary
Docker Configuration `Dockerfile`	Added `ARG` and `ENV` for `NEMOCLAW_CONTEXT_WINDOW`, `NEMOCLAW_MAX_TOKENS`, `NEMOCLAW_REASONING`. Python config generation now reads `context_window` and `max_tokens` from env (with defaults) and converts `NEMOCLAW_REASONING` from string to boolean; replaced prior hard-coded values in generated `openclaw` config.
Dockerfile Patching Logic `src/lib/onboard.ts`	`patchStagedDockerfile()` reads host `process.env` for the three NEMOCLAW vars and conditionally replaces corresponding `ARG` default lines in staged Dockerfiles. Validates context/max as positive integers and reasoning as exact `"true"`/`"false"`; leaves ARG lines unchanged on invalid or missing env.
Tests `test/onboard.test.ts`	Added test that creates a temp staged `Dockerfile`, sets `process.env` values for the three vars, runs `patchStagedDockerfile(...)`, asserts ARG lines reflect env-provided values (including converting `"true"` to `ARG ...=true`), and restores env/cleans up in `finally`.

Sequence Diagram(s)

sequenceDiagram
    participant Host as Host (process.env)
    participant Patcher as onboard.patchStagedDockerfile
    participant Dockerfile as Staged Dockerfile
    participant Builder as Image build / Python config generator

    Host->>Patcher: provide NEMOCLAW_* env vars
    Patcher->>Dockerfile: read & validate ARG lines
    Patcher->>Dockerfile: replace ARG defaults when valid
    Dockerfile->>Builder: build image with updated ENV
    Builder->>Builder: Python reads env, computes contextWindow/maxTokens/reasoning
    Builder->>GeneratedConfig: write models with computed values

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

🐇 I nibbled ARGs in a Dockerfile glade,
Tuned context and tokens, then bravely I bade,
"Reason true or false, let the builder decide,"
I hopped, I patched, and watched configs abide. 🥕✨

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: propagating context window, max tokens, and reasoning settings to the sandbox image configuration across all modified files.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

coderabbitai

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/lib/onboard.ts`:
- Around line 1027-1040: POSITIVE_INT_RE currently permits "0", so update the
validation used for contextWindow and maxTokens to only accept positive integers
> 0 (e.g., change POSITIVE_INT_RE to a regex that requires a non-zero leading
digit like /^[1-9][0-9]*$/ or replace the regex check with numeric parsing and a
> 0 check); ensure the same updated validation is applied before calling
dockerfile.replace for NEMOCLAW_CONTEXT_WINDOW and NEMOCLAW_MAX_TOKENS so zero
values are rejected.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 3b708133-f82e-494c-9ea6-bfbead0d0142

📥 Commits

Reviewing files that changed from the base of the PR and between ec4d058 and cfe4d69.

📒 Files selected for processing (3)

Dockerfile
src/lib/onboard.ts
test/onboard.test.ts

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

wscurran · 2026-04-16T15:20:43Z

✨ Thanks for submitting this PR that proposes a fix to propagate context window, max tokens, and reasoning to the sandbox image, which could help improve the functionality of NemoClaw.

coderabbitai

🧹 Nitpick comments (1)

src/lib/onboard.ts (1)
1231-1248: Add negative-path tests for env validation edge cases.

Current coverage demonstrates the happy path, but this block would benefit from explicit tests for invalid values ("0", negatives, non-numeric, whitespace-padded, "TRUE"), asserting the original ARG defaults remain unchanged.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@src/lib/onboard.ts` around lines 1231 - 1248, Add negative-path unit tests
that exercise the env validation around POSITIVE_INT_RE and the reasoning check:
set process.env.NEMOCLAW_CONTEXT_WINDOW and NEMOCLAW_MAX_TOKENS to invalid
values like "0", "-1", "abc", " 5 ", and "TRUE", then call the function that
produces/returns the dockerfile (the code block using POSITIVE_INT_RE and
NEMOCLAW_REASONING) and assert the original ARG lines for
NEMOCLAW_CONTEXT_WINDOW and NEMOCLAW_MAX_TOKENS (and the reasoning ARG) remain
unchanged; ensure tests restore env state after each case and include separate
cases for each invalid input to verify the replacements do not occur.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@src/lib/onboard.ts`:
- Around line 1231-1248: Add negative-path unit tests that exercise the env
validation around POSITIVE_INT_RE and the reasoning check: set
process.env.NEMOCLAW_CONTEXT_WINDOW and NEMOCLAW_MAX_TOKENS to invalid values
like "0", "-1", "abc", " 5 ", and "TRUE", then call the function that
produces/returns the dockerfile (the code block using POSITIVE_INT_RE and
NEMOCLAW_REASONING) and assert the original ARG lines for
NEMOCLAW_CONTEXT_WINDOW and NEMOCLAW_MAX_TOKENS (and the reasoning ARG) remain
unchanged; ensure tests restore env state after each case and include separate
cases for each invalid input to verify the replacements do not occur.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ae6a0546-080b-493e-8ad3-c789e33a579c

📥 Commits

Reviewing files that changed from the base of the PR and between ea0d3cf and 8ef8ac2.

📒 Files selected for processing (3)

Dockerfile
src/lib/onboard.ts
test/onboard.test.ts

✅ Files skipped from review due to trivial changes (1)

test/onboard.test.ts

🚧 Files skipped from review as they are similar to previous changes (1)

Dockerfile

ericksoa

Reviewed from a security perspective — input validation is solid, no injection vectors, trust boundary is appropriate. LGTM.

…d reasoning env vars (#2108) ## Summary Documents the `NEMOCLAW_CONTEXT_WINDOW`, `NEMOCLAW_MAX_TOKENS`, and `NEMOCLAW_REASONING` build-time environment variables that landed in #1956. These variables now propagate through the Dockerfile ARG chain into `openclaw.json`, but were not mentioned anywhere in the published docs. ## Changes - Add a "Tune Model Metadata" section to `docs/inference/switch-inference-providers.md` with a variable reference table, defaults, value constraints, an example, and the `nemoclaw onboard --resume --recreate-sandbox` flow for applying changes to existing sandboxes. - Regenerate `.agents/skills/nemoclaw-user-configure-inference/SKILL.md` from the updated source. - Sync `.agents/skills/nemoclaw-user-reference/references/troubleshooting.md` with the current `docs/reference/troubleshooting.md` (brings the autogen reference back in line with the `openclaw update` entry added in #2033). ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [x] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Verification - [x] `npx prek run --all-files` passes - [ ] `npm test` passes - [ ] Tests added or updated for new or changed behavior - [x] No secrets, API keys, or credentials committed - [x] Docs updated for user-facing behavior changes - [x] `make docs` builds without warnings (doc changes only) - [x] Doc pages follow the [style guide](https://github.com/NVIDIA/NemoClaw/blob/main/docs/CONTRIBUTING.md) (doc changes only) - [ ] New doc pages include SPDX header and frontmatter (new pages only) ## AI Disclosure - [x] AI-assisted — tool: Cursor --- Signed-off-by: Miyoung Choi <miyoungc@nvidia.com> Made with [Cursor](https://cursor.com)  ## Summary by CodeRabbit * **Documentation** * Added guide for tuning model metadata via build-time configuration variables. * Documented three new parameters for context window, max tokens, and reasoning, including accepted values, defaults, and that invalid values are ignored. * Clarified that changes apply during image build and require recreating a sandbox to take effect. * Simplified update-timeout troubleshooting with a single rebuild workflow. * Updated documentation versioning to include the new release.  --------- Signed-off-by: Miyoung Choi <miyoungc@nvidia.com>

…son (#2441)  ## Summary The `NEMOCLAW_INFERENCE_INPUTS` variable is not sent all the way to the `openclaw.json` file (and thus the agent configuration). The `input` field in `models[0]` is hardcoded to `['text']` in the Dockerfile, so users who select a vision-capable model via `NEMOCLAW_MODEL` cannot tell OpenClaw that the model accepts image input. This extends #1956's plumbing pattern to cover the last remaining hardcoded inference field. ## Related Issue Fixes #2421 ## Changes - adds 1 ARG to the `Dockerfile` with default `text` (matching the previously-hardcoded value), promotes it to ENV, and the Python script reads from `os.environ` . - `patchStagedDockerfile()` in `onboard.ts` reads `NEMOCLAW_INFERENCE_INPUTS` from the host and patches the ARG line before image build, validated by regex `/^(text|image)(,(text|image))*$/` matching OpenClaw's `model.input` schema (`z.array(z.union([z.literal("text"), z.literal("image")]))`) - Unit tests to verify: positive case (`text,image` → baked) plus a 6-case reject table (`undefined / audio / text, / Text,Image / text, image / shell-injection attempt`) ## Type of Change - [ √] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Verification  - [ √] `npx prek run --all-files` passes - [ √] `npm test` passes - [ √] Tests added or updated for new or changed behavior - [√ ] No secrets, API keys, or credentials committed - [ ] Docs updated for user-facing behavior changes - [ ] `make docs` builds without warnings (doc changes only) - [ ] Doc pages follow the [style guide](https://github.com/NVIDIA/NemoClaw/blob/main/docs/CONTRIBUTING.md) (doc changes only) - [ ] New doc pages include SPDX header and frontmatter (new pages only) ## AI Disclosure - [ ] AI-assisted — tool: Cursor ---  Signed-off-by: rluo8 <ruluo@nvidia.com>  ## Summary by CodeRabbit * **New Features** * Added build/runtime environment support to specify model input modalities (e.g., text, image) with a safe default of "text", enabling image-capable modal setups via env configuration. * Inputs are validated and parsed into a trimmed, comma-separated list before use. * **Tests** * Added regression tests for parsing and validation of the input-modality setting, covering valid multi-value inputs and malformed or injection-like values.  --------- Signed-off-by: Rui Luo <ruluo@nvidia.com>

…son (NVIDIA#2441)  ## Summary The `NEMOCLAW_INFERENCE_INPUTS` variable is not sent all the way to the `openclaw.json` file (and thus the agent configuration). The `input` field in `models[0]` is hardcoded to `['text']` in the Dockerfile, so users who select a vision-capable model via `NEMOCLAW_MODEL` cannot tell OpenClaw that the model accepts image input. This extends NVIDIA#1956's plumbing pattern to cover the last remaining hardcoded inference field. ## Related Issue Fixes NVIDIA#2421 ## Changes - adds 1 ARG to the `Dockerfile` with default `text` (matching the previously-hardcoded value), promotes it to ENV, and the Python script reads from `os.environ` . - `patchStagedDockerfile()` in `onboard.ts` reads `NEMOCLAW_INFERENCE_INPUTS` from the host and patches the ARG line before image build, validated by regex `/^(text|image)(,(text|image))*$/` matching OpenClaw's `model.input` schema (`z.array(z.union([z.literal("text"), z.literal("image")]))`) - Unit tests to verify: positive case (`text,image` → baked) plus a 6-case reject table (`undefined / audio / text, / Text,Image / text, image / shell-injection attempt`) ## Type of Change - [ √] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Verification  - [ √] `npx prek run --all-files` passes - [ √] `npm test` passes - [ √] Tests added or updated for new or changed behavior - [√ ] No secrets, API keys, or credentials committed - [ ] Docs updated for user-facing behavior changes - [ ] `make docs` builds without warnings (doc changes only) - [ ] Doc pages follow the [style guide](https://github.com/NVIDIA/NemoClaw/blob/main/docs/CONTRIBUTING.md) (doc changes only) - [ ] New doc pages include SPDX header and frontmatter (new pages only) ## AI Disclosure - [ ] AI-assisted — tool: Cursor ---  Signed-off-by: rluo8 <ruluo@nvidia.com>  ## Summary by CodeRabbit * **New Features** * Added build/runtime environment support to specify model input modalities (e.g., text, image) with a safe default of "text", enabling image-capable modal setups via env configuration. * Inputs are validated and parsed into a trimmed, comma-separated list before use. * **Tests** * Added regression tests for parsing and validation of the input-modality setting, covering valid multi-value inputs and malformed or injection-like values.  --------- Signed-off-by: Rui Luo <ruluo@nvidia.com>

fix(onboard): propagate context window, max tokens, and reasoning to …

cfe4d69

…sandbox image

coderabbitai Bot reviewed Apr 16, 2026

View reviewed changes

Comment thread src/lib/onboard.ts Outdated

Update src/lib/onboard.ts per CodeRabbit suggestion

ea0d3cf

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

wscurran added the fix label Apr 16, 2026

wscurran added the NemoClaw CLI label Apr 16, 2026

jyaunches mentioned this pull request Apr 16, 2026

fix(onboard): propagate rotated messaging credentials to sandbox L7 proxy #1967

Merged

6 tasks

Merge branch 'main' into fix-context-window

8ef8ac2

coderabbitai Bot reviewed Apr 18, 2026

View reviewed changes

ericksoa approved these changes Apr 18, 2026

View reviewed changes

ericksoa merged commit 56af860 into NVIDIA:main Apr 18, 2026
1 check passed

javlund deleted the fix-context-window branch April 18, 2026 06:29

miyoungc mentioned this pull request Apr 20, 2026

docs: refresh for v0.0.20 and document context window, max tokens, and reasoning env vars #2108

Merged

13 tasks

rluo8 mentioned this pull request Apr 24, 2026

fix(onboard): propagate NEMOCLAW_INFERENCE_INPUTS to baked openclaw.json #2441

Merged

8 tasks

wscurran added area: cli Command line interface, flags, terminal UX, or output bug-fix PR fixes a bug or regression and removed NemoClaw CLI labels Jun 3, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(onboard): propagate context window, max tokens, and reasoning to sandbox image#1956

fix(onboard): propagate context window, max tokens, and reasoning to sandbox image#1956
ericksoa merged 3 commits into
NVIDIA:mainfrom
javlund:fix-context-window

javlund commented Apr 16, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Apr 16, 2026 •

edited

Loading

Walkthrough

Changes

Sequence Diagram(s)

Estimated code review effort

Poem

❌ Failed checks (1 warning)

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

wscurran commented Apr 16, 2026

Uh oh!

coderabbitai Bot left a comment

Uh oh!

ericksoa left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

javlund commented Apr 16, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Type of Change

Verification

AI Disclosure

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Apr 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Sequence Diagram(s)

Estimated code review effort

Poem

❌ Failed checks (1 warning)

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

wscurran commented Apr 16, 2026

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

ericksoa left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

javlund commented Apr 16, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Apr 16, 2026 •

edited

Loading