test(e2e): exercise model override startup path

[ericksoa]

Outcome

The image-level model-override test now exercises the real container entrypoint and proves the complete startup transition: JSON rewrite, refreshed hash, sandbox ownership, and mutable modes.

Reason

PR #11001 repaired the deterministic current-main normalize_mutable_config_perms: command not found harness failure with local stubs. Its exact-head Delivery and Verification Advisors correctly found that those stubs bypassed production permission normalization, owner dispatch, and hash refresh. #11001 was merged externally before this correction could be published.

Related issues

Follow-up to #11001.

Changes

• Remove the brittle extraction of apply_model_override and all three local helper stubs.
• Run the built image through its real root entrypoint with NEMOCLAW_MODEL_OVERRIDE.
• Require the one-shot command to observe the rewritten primary and model entries.
• Require .config-hash to validate and the mutable directory, config, and hash to retain their sandbox ownership and expected modes.
• Fail the inner command immediately if any JSON, hash, ownership, or mode check fails, before the success marker can be emitted.

Verification

• shfmt — passed.
• shellcheck — passed.
• Codebase growth guardrail and repository hooks — passed.
• git diff --check origin/main...HEAD — passed.
• Exact-head managed image E2E on test(e2e): reduce Hermes Discord assertions #11001 proved the original 44/45 failure was resolved by the predecessor fixture update; this stronger real-entrypoint form requires its own CI run before merge.
• Diff inspection — no secrets, API keys, or credentials.

Review notes

This is the already-prepared correction for the exact-head Advisor findings on #11001. It replaces the merged stubs rather than layering another test path. The first #11003 Advisor pass found that the new inner shell needed fail-fast behavior; the current head adds set -e before every required check.

---

Signed-off-by: Aaron Erickson aerickson@nvidia.com

Summary by CodeRabbit

• Tests
  • Expanded end-to-end coverage for the model-override scenario.
  • Added validation that configuration changes are correctly applied and remain consistent.
  • Added checks for configuration integrity and expected file ownership and permission settings.
  • Updated test execution to stop immediately when a validation step fails.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 5c46e0a9-a8e1-4699-98ec-6c04b6c994d1

📥 Commits

Reviewing files that changed from the base of the PR and between 968a039 and 9adb50f.

📒 Files selected for processing (1)

• test/e2e-gateway-isolation.sh

Included review availability: Your plan provides up to 12 included reviews per hour; 5 remain after this review.

---

📝 Walkthrough

Walkthrough

The model-override end-to-end test now uses the image entrypoint with fail-fast behavior. It validates the overridden model data, configuration hash, and ownership and permission modes for mutable configuration paths.

Changes

Gateway override validation

Layer / File(s)	Summary
Entrypoint override and configuration checks test/e2e-gateway-isolation.sh	Test 27 runs through the image entrypoint, enables fail-fast behavior, and validates the model override, config hash, ownership, and mutable permission modes.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 9adb5

This updates the end-to-end model-override test to exercise the normal container entrypoint and validate configuration integrity and permissions. No current merge-blocking risk remains.

Suggested reviewers: aasthajh

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: the E2E test now exercises the model-override startup path.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 1…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

Full details: Docstring Coverage

Explanation

No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 1 files.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/model-override-startup-e2e

---

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

[github-code-quality]

Code Coverage Overview

Languages: TypeScript

TypeScript / code-coverage/plugin

The overall line coverage in commit 9adb50f in the fix/model-override-s... branch remains at 96%, unchanged from commit 3d75441 in the main branch.

---

_{Updated September 03, 2026 21:26 UTC}

[rsliter]

Approved at 9adb50f.

I reviewed the complete one-file diff and the production entrypoint path. Test 27 now exercises the real root entrypoint and owner dispatch, fails closed before emitting its success marker, and validates the rewritten JSON, refreshed config hash, and sandbox ownership and modes. The exact-head gateway-isolation E2E passed, as did static checks, ShellCheck, CodeQL, DCO, and signed-commit verification. CodeRabbit reported no actionable findings. The completed Advisor migration and documentation reviews found no issues. Security review found no new credential, dependency, authorization, input, or configuration exposure; this strengthens coverage of the existing privilege and integrity boundary.

The failed statuses are infrastructure or aggregate fallout, not candidate-owned failures. Four Advisor specialists report transport errors with TCP connection refused, then produce no review artifacts. cli-test-shards (8) was canceled during Vitest with only "The operation was canceled" and no test failure. cli-tests and checks fail only because that shard did not complete. The same base commit, 3d75441, passed CI / Main Branch, including shard 8, cli-tests, and checks. The changed shell E2E is not part of that CLI shard, and its exact-head gateway-isolation job passed.