fix(e2e): restore messaging proxy reachability

[jyaunches]

Outcome

The trusted messaging-providers target can expose its credential-free fake API proxy on the OpenShell bridge while keeping the credential-bearing fake API on an internal-only network. The proxy remains unable to use masqueraded outbound egress.

Reason

The latest completed target evidence for the superseded #10628 failed the installed WeChat runtime with ECONNREFUSED 172.18.0.1:32771. The host port was not reachable from the OpenShell sandbox: the main implementation bound it only to host loopback, while the attempted internal-only replacement could not install usable published-port forwarding.

Related issues

• Supersedes fix(e2e): restore fake API port publishing #10628.
• Relates to the messaging-providers failure in full main E2E run 33350578179, job 99363408336.

Changes

• Inspect and validate the private IPv4 gateway of the configured OpenShell Docker bridge before creating fake API resources.
• Keep the fake API on its internal-only network with no published ports.
• Create a dedicated non-masquerading bridge for the credential-free proxy, bind its ephemeral ports only to the OpenShell gateway, connect it to the internal API network while stopped, and then start it.
• Verify Docker reports the exact gateway binding and retain reverse-order container and network cleanup.
• Extend deterministic support coverage for the topology, credential separation, cleanup order, invalid network names, ambiguous bridge inspection, and incorrect host bindings.

Verification

• npx vitest run --project e2e-support test/e2e/support/messaging-providers-runtime-proofs.test.ts — 30 tests passed.
• npx vitest run --project integration test/automation/pull-requests/growth-guardrails.test.ts — 33 tests passed.
• npm run test:e2e-phases:check — 132 tests across 90 files passed semantic phase coverage.
• npm run checks:repository — repository checks passed.
• npm run test:titles:check — test title check passed.
• npx oxfmt --check test/e2e/live/messaging-providers-helpers.ts test/e2e/support/messaging-providers-runtime-proofs.test.ts — formatting passed.
• Pre-commit and pre-push hooks — passed, including TypeScript, repository, secret, E2E phase, and growth checks.
• Diff review — no secrets, API keys, or credentials are present.

Review notes

E2E root cause

• Source evidence: manual PR E2E run 33507949591 tested superseded candidate db5a9c6793accb0bd9167e9421becdf03940ad39 against base 9169ab0790d9a1be68b439731c01b696f75ff085.
• Stable signature: the installed WeChat runtime failed with ECONNREFUSED 172.18.0.1:32771 after publication and startup traffic proof had passed.
• Cleanup evidence: all 12 registered cleanup entries passed with no failures.
• Focused scope: repair only the fake API proxy publication topology and its deterministic support proof.
• Required external evidence: one trusted latest-commit messaging-providers target run with complete cleanup proof. This PR remains draft until that target and exact-commit CI and review gates are green.

Security boundary

The fake API alone receives provider credentials and has no host publication. The host-published proxy receives no provider credentials, drops all capabilities, uses a read-only filesystem, and attaches to a separate bridge with IP masquerading disabled. Docker network names and the OpenShell bridge address fail closed before resources are created.

---

Signed-off-by: Julie Yaunches jyaunches@nvidia.com

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

_{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 8c5421a in the codex/fix-messaging-... branch remains at 96%, unchanged from commit 95c0a60 in the main branch.

TypeScript / code-coverage/cli

The overall line coverage in commit 8c5421a in the codex/fix-messaging-... branch remains at 84%, unchanged from commit 95c0a60 in the main branch.

Show a line coverage summary of the most impacted files.

File	main 95c0a60	codex/fix-messaging-... 8c5421a	+/-
src/lib/sandbox...rce-identity.ts	82%	82%	0%
src/lib/onboard...uild-context.ts	74%	75%	+1%

---

_{Updated September 02, 2026 00:34 UTC}

[jyaunches]

Trusted target reconciliation for exact head 781b13c7bfde09e37d903270c2abc3c4dddba631:

• Run: https://github.com/NVIDIA/NemoClaw/actions/runs/33517907472 (attempt 1)
• Correlation: e2f313a9-86a3-4ad0-906e-c9382bc9c46d
• Receipt: candidate 781b13c7bfde09e37d903270c2abc3c4dddba631, base/workflow 9169ab0790d9a1be68b439731c01b696f75ff085, target messaging-providers
• Result: failed before the changed proxy-topology boundary. Onboarding's hosted Chat Completions probe timed out on every bounded attempt; the redacted summary is Chat Completions API: curl exit 28.
• Classification: external inference-provider/network timing failure, not candidate-attributable. This PR changes only the messaging live helper and deterministic support proof; the run stopped during install/onboarding in phase 2. Earlier run 33507949591 on the same protected base completed onboarding and reached phase 6.
• Cleanup: all 3 registered entries passed; failures=[].
• Risk signal: testedSha=expectedSha=781b13c7..., failed=1, unhandledErrors=0.

This run is historical infrastructure evidence only. It is not a qualifying green latest-head target result. The PR remains draft; no rerun or replacement dispatch was made.

[github-actions]

PR Review Advisor finished for commit 5227d6d. Include the Advisor findings in the complete PR feedback collection. Verify and group valid findings before repair.

All previous runs