fix(e2e): echo unique launch replies in mock inference

[prekshivyas]

Summary

The release E2E run showed two agent-turn failures: OpenClaw rendered the expected unique reply inside terminal control sequences, and the shared mock inference endpoint returned fixed PONG to Hermes. Current main already contains the OpenClaw OSC normalization and regression from #9034; this change repairs the remaining mock behavior by deriving each unique reply from the exact launch prompt.

Related Issue

Fixes #9046

Changes

• Add an opt-in launch-prompt parser to the fake Chat Completions endpoint. The shared fake server serves unrelated fixtures, so changing its default response would alter other tests; the adapter enables the parser only for mock inference.
• Accept only the generated four-fragment prompt and preserve the configured PONG response for partial or altered prompts.
• Add a regression test that failed with fixed PONG before the change and now verifies both the unique reply and fallback behavior.
• Close the detection gap where the adapter test exercised only a fixed PONG prompt and did not cover the generated launch-turn contract.

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)

Quality Gates

• Tests added or updated for changed behavior
• Existing tests cover changed behavior — justification:
• Tests not applicable — justification:
• Docs updated for user-facing behavior changes
• Docs not applicable — justification: This changes an internal E2E mock endpoint and its support test. It does not change a public command, API, configuration, default, workflow, or supported product behavior.
• Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging)
• Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: The mock Chat Completions input boundary accepts one exact uppercase prompt shape. A negative regression preserves the fixed response for altered input. Authentication, credential redaction, request metadata, production inference routes, dependencies, and network policy are unchanged.
• Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue:

Documentation Writer Review

• Documentation writer subagent reviewed the completed changes
• Result: no-docs-needed
• Evidence: The reviewed four-file diff changes only E2E fixture and support behavior. It adds no public API, CLI, configuration, default, workflow, or supported product behavior.
• Agent: Codex Desktop

DGX Station Hardware Evidence

• Tested on DGX Station
• Tested commit:
• Station profile/scenario:
• Result:
• Supporting evidence:

Verification

• PR description includes a Signed-off-by: line and every commit appears as Verified in GitHub
• Normal pre-commit, commit-msg, and pre-push hooks passed, or npm run validate:pr passed after refreshing origin/main when hooks were skipped or unavailable
• Targeted behavior tests pass for the current change set, or tests are marked not applicable above — npx vitest run --project e2e-support test/e2e/support/inference-adapter.test.ts test/e2e/support/hosted-inference.test.ts test/e2e/support/device-auth-health-helpers.test.ts test/e2e/support/launch-agent-turn.test.ts: 4 files passed, 32 tests passed, 1 platform skip.
• Applicable broad gate passed — npm test for broad runtime/test-harness changes; npm run check for repo-wide validation/coverage changes — command/result: Not applicable to this isolated mock behavior. An additional npx vitest run --project e2e-support passed 197 files and failed in seven unrelated macOS host-sensitive files that require GNU find, trusted Homebrew state, systemd or Ollama, or longer subprocess timeouts.
• Quality Gates section completed with required justifications or waivers
• No secrets, API keys, or credentials committed
• npm run 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)

---

Signed-off-by: Prekshi Vyas prekshiv@nvidia.com

Summary by CodeRabbit

• Tests
  • Enhanced end-to-end inference testing for launch-specific prompts.
  • Added coverage for both streaming and non-streaming responses.
  • Verified exact prompt-derived responses while preserving default responses for similar prompts without explicit instructions.

[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]

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: a348cabc-39e5-4adb-a1d0-a6777479afc1

📥 Commits

Reviewing files that changed from the base of the PR and between ddcdd56 and 5d38dbe.

📒 Files selected for processing (4)

• test/e2e/fixtures/fake-openai-compatible.ts
• test/e2e/fixtures/inference-adapter.ts
• test/e2e/lib/fake-openai-compatible-api.mts
• test/e2e/support/inference-adapter.test.ts

---

📝 Walkthrough

Walkthrough

The fake inference server now supports opt-in launch replies derived from prompt fragments. The inference adapter enables this mode, and end-to-end tests verify exact launch responses while preserving PONG for non-matching prompts.

Changes

Prompt-derived launch reply flow

Layer / File(s)	Summary
Server configuration test/e2e/fixtures/fake-openai-compatible.ts, test/e2e/fixtures/inference-adapter.ts, test/e2e/lib/fake-openai-compatible-api.mts	The fixtures expose and propagate launchReplyFromPrompt. The fake server reads the setting from NEMOCLAW_FAKE_OPENAI_LAUNCH_REPLY_FROM_PROMPT.
Prompt response handling and validation test/e2e/lib/fake-openai-compatible-api.mts, test/e2e/support/inference-adapter.test.ts	Matching prompts produce the constructed launch reply for streaming and non-streaming completions. Non-matching prompts retain PONG. Tests verify both cases.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Mergeability Score: ⚪ Minimal · up to 5d38d

This changes only opt-in E2E mock response behavior, with targeted tests passing; no actionable merge-blocking risk remains.

Suggested reviewers: cv

Sequence Diagram(s)

sequenceDiagram
  participant InferenceAdapterTest
  participant InferenceAdapter
  participant FakeOpenAiCompatibleServer
  InferenceAdapterTest->>InferenceAdapter: send launch-specific prompt
  InferenceAdapter->>FakeOpenAiCompatibleServer: request chat completion
  FakeOpenAiCompatibleServer->>FakeOpenAiCompatibleServer: validate prompt fragments
  FakeOpenAiCompatibleServer-->>InferenceAdapter: constructed launch reply
  InferenceAdapter-->>InferenceAdapterTest: exact response

Loading

🚥 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 describes the primary change: mock inference now echoes unique launch replies.
Linked Issues check	✅ Passed	The PR updates fake inference to echo prompt-encoded unique replies, preserves PONG fallback, and adds regression coverage for issue #9046.
Out of Scope Changes check	✅ Passed	All changes are limited to E2E fixtures, fake inference behavior, and tests directly related to issue #9046.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ 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 agent/normalize-agent-turn-output

---

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

[github-code-quality]

Code Coverage Overview

Languages: TypeScript

TypeScript / code-coverage/plugin

The overall coverage in commit 5d38dbe in the agent/normalize-agen... branch remains at 96%, unchanged from commit ddcdd56 in the main branch.

[github-actions]

PR Review Advisor — No blocking findings reported

Advisor assessment: No blocking advisor findings reported
Next action: No advisor follow-up needed.
Findings: 0 blockers · 0 warnings · 0 suggestions

Model lanes

• GPT-5.6 Terra (primary): Completed · high confidence · 0 blockers · 0 warnings · 0 suggestions
• Nemotron 3 Ultra (second opinion): Completed · high confidence · 0 blockers · 0 warnings · 0 suggestions
• Model comparison: normalized findings match; normalized terminology decisions differ; normalized E2E selections differ; severity counts match.

2 terminology differences from the second opinion

Advisory only. These are normalized differences from the primary terminology receipt.

• mock prompt at test/e2e/support/inference-adapter.test.ts:183: selected only by the second-opinion lane as conflict.
• unique launch reply at test/e2e/support/inference-adapter.test.ts:183: selected only by the second-opinion lane as justified.

3 additional E2E selections from the second opinion

Advisory only. The primary lane did not select these E2E jobs or targets.

• e2e-all: The completed second-opinion lane identified E2E coverage that the primary lane omitted.
• hermes-inference-switch: The completed second-opinion lane identified E2E coverage that the primary lane omitted.
• openclaw-inference-switch: The completed second-opinion lane identified E2E coverage that the primary lane omitted.

Second-opinion terminology and E2E selections are advisory. Live E2E does not run automatically for pull requests.

1 semantic terminology decision

Terminology decisions are advisory. They affect the assessment only when a separate finding identifies concrete semantic impact.

• justified — launch reply at test/e2e/support/inference-adapter.test.ts:183: Keep “launch reply” for the prompt-derived reply used by launch-turn tests.

E2E guidance

Advisory only. A maintainer can dispatch the default E2E suite for the commit under review.

Recommended E2E: inference-routing

Manual-only E2E: cloud-onboard, security-posture, cloud-inference, network-policy
The manual PR workflow does not run these selectors for the commit under review. Run them from reviewed code on main.

1 optional E2E recommendation

• hermes-e2e

Workflow run details

This automated review informs maintainers. Warnings and suggestions do not require a response. A maintainer decides whether to merge.