test(server): record the Codex thread rollback fixture live on 0.156.1

[juliusmarminge]

The Codex thread_rollback transcript was captured on Codex 0.120 and its rollback exchange was hand-edited into the 0.156 thread/read + thread/turns/list + thread/revert shape, so nothing proved the revert path against a real app-server. This re-records it live on Codex 0.156.1.

What changed

• Fixture: thread_rollback/codex_transcript.ndjson is now a straight recording from Codex 0.156.1. Every frame is live and nothing was hand-edited.

• Recorder (record-codex-app-server-replay-fixture.ts):

  • sends the adapter's initialize notification opt-out and turn/start defaults (never-approve, full-access sandbox, cwd, model, detailed summary, user reviewer);
  • reads the thread before reverting, the same way the adapter checks the history mode;
  • takes --model / T3_CODEX_REPLAY_MODEL;
  • writes the recording cwd as <workspace> in outbound frames (codexReplayRecordingRecords.ts).

  Together these make the transcript replay against the adapter without edits.

• Model: the fixture's Codex variant gets its own modelSelection (gpt-5.6-luna), because the hub no longer serves gpt-5.4. The shared CODEX_MODEL_SELECTION is unchanged.

• Assertions (codex_output.ts):

  • check that the answer after the rollback ("repeat the conversation") includes the first turn and not the reverted second one, which proves Codex's own context was rewound;
  • accept the two reasoning items the current model emits.

What the live recording showed

The recording agrees with the adapter's assumptions; nothing contradicts them:

• thread/start returns historyMode: "paginated" with no historyMode param, so threads T3 starts are revertable.
• thread/turns/list (descending, limit 1) returns the second turn, and thread/revert {beforeTurnId} succeeds. Its response carries thread (with turns: []), turnsBackwardsCursor and itemsBackwardsCursor.
• During the revert, Codex unloads and reloads the thread. It emits thread/status/changed notLoaded → idle, then sends thread/reverted after the response.
• The recording also has a model-mismatch warning during the reload. The recorder causes it, not Codex: it sends thread/start with {}, so the thread starts on the hub's default model, and the revert reloads it with a mismatch warning. The adapter sends model on thread/start, so users likely don't see it. A follow-up will make the recorder send model/cwd on thread/start.
• The next turn's "repeat the conversation" answer only contains the first exchange.

Verification

• vp test run src/orchestration-v2/testkit/OrchestratorReplayFixtures.integration.test.ts -t thread_rollback: 2 passed (Codex and Claude).
• vp test run on CodexReplayFixtures, ThreadFork, ThreadMergeBack, OrchestratorReplayFixtures.contract, CodexAdapterV2.test.ts and scripts/codexReplayRecordingRecords.test.ts: 139 passed.
• vp exec tsc --noEmit -p . in apps/server: no errors or warnings. vp lint and vp fmt on the touched files are clean.
• vp run knip:check passes.
• Recorded with T3_CODEX_BIN=<codex 0.156.1> node scripts/record-codex-app-server-replay-fixture.ts --scenario thread_rollback --model gpt-5.6-luna.
• Not run: the other Codex fixtures, which still pin gpt-5.4 and pre-0.156 recordings.

Model: Claude Opus 5.5 (Claude Code)

🤖 Generated with Claude Code

---

[macroscopeapp]

Approvability

Verdict: Approved at 0cc5e27

Macroscope's review found this PR approvable — This PR refreshes a Codex rollback replay fixture and its development-only recording harness, adding assertions and workspace normalization without changing production request-path behavior. The large diff is primarily regenerated test transcript data.

No code changes detected at e37e762. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.4 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	2	—	8	✅
Claude	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	20.8 KiB	—	29.3 KiB	✅
Claude	Live turn messages	—	2	—	8	✅

Baseline: unavailable · PR result: e37e762 · Source CI: failure

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.