perf(server): Codex plan streaming no longer commits every token

[t3dotgg]

Closes #18956.

When Codex streams a proposed plan, each token delta commits three events (node.updated, plan.updated, turn-item.updated), and two of them carry the whole plan written so far. The recorded 1.3 KB plan fixture commits 782 events (2.1 MB). Assistant text and reasoning avoid this because they go through makeProviderTextDeltaCoalescer. Plan deltas did not.

Fix

item/plan/delta now appends to a makeProviderTextDeltaCoalescer, the same one the Codex adapter uses for assistant text and reasoning. It flushes at most every 400 ms, the cadence clients already see for assistant text.

• item/completed completes the plan through the coalescer. It drops any pending flush and always emits the final plan, as before.
• When the turn ends, any buffered plan text is flushed. On an interrupted turn the partial plan stays streaming, as it did before.
• No contract or wire change. Clients still see the plan stream in, a few times a second.

Proof

Issue method: the recorded proposed_plan Codex fixture replayed through the orchestrator, counting result.domainEvents:

	events	bytes	turn-item.updated	plan.updated	node.updated
main	782	2,110,566	254	252	256
this PR	29	70,780	3	1	5

That fixture has no timing, so all deltas arrive inside one flush window. To show timed streaming, the new test streams a proposed plan a few times a second instead of once per delta (CodexAdapterV2.test.ts) sends 100 deltas 20 ms apart:

	provider events	plan.updated	bytes
main (test fails)	306	101	298,375
this PR	18	5	15,625

It also checks that the final plan has the full text and that only the last update has streaming: false. A live 5.3 s plan should take about 14 flushes, so roughly 45 events instead of 782.

• vp test run on CodexAdapterV2.test.ts, CodexReplayFixtures.integration.test.ts, OrchestratorReplayFixtures.integration.test.ts, OrchestratorReplayFixtures.contract.test.ts: 311 passed.
• Server tsc --noEmit clean. Scoped lint and format clean.

Not verified: a live Codex plan-mode run. Other providers already batch plan updates. Cursor, for example, commits 4 plan.updated events for the same fixture.

Created with Claude Opus 5.5 in Claude Code (T3 Code).

🤖 Generated with Claude Code

[macroscopeapp]

Approvability

Verdict: Approved at 82f35c8

Macroscope's review found this PR approvable — This is a focused Codex streaming performance fix that reduces redundant persistence writes while preserving final plan content, lifecycle state, and event shapes. Targeted replay coverage verifies coalesced updates and complete final output.

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

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Path: .coderabbit.config.ts
• Review profile: CHILL
• Plan: Team
• Run ID: 7c060308-23d2-49b2-b283-2b61445401b1

📥 Commits

Reviewing files that changed from the base of the PR and between acb9027 and 82f35c8.

📒 Files selected for processing: 2

• apps/server/src/orchestration-v2/Adapters/CodexAdapterV2.test.ts
• apps/server/src/orchestration-v2/Adapters/CodexAdapterV2.ts

Limit details: You’ve used all 10 included reviews currently available. Your 14 included PR review attempts over the past 7 days set your current allowance at 10 reviews per hour.

📝 Walkthrough

Codex proposed-plan deltas are now coalesced before artifact updates are emitted. The adapter flushes buffered text at completion and turn finalization. A replay test checks update counts, final markdown, and streaming status.

Changes: Codex Proposed-Plan Streaming

Layer / File(s)	Summary
Coalesced plan projection apps/server/src/orchestration-v2/Adapters/CodexAdapterV2.ts	The adapter uses a 400 ms flush interval and a coalesced text stream to build and emit proposed-plan artifacts. It removes the per-item string map.
Plan stream lifecycle and replay validation apps/server/src/orchestration-v2/Adapters/CodexAdapterV2.ts, apps/server/src/orchestration-v2/Adapters/CodexAdapterV2.test.ts	Plan notifications append text to the coalescer. Plan completion and turn finalization flush buffered text. The replay test checks coalesced update counts, complete markdown, and final streaming status.

Priority: ➖ Normal

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

Change: Bug fix

Suggested reviewers: juliusmarminge

Merge Risk: ⚪ Minimal · up to 82f35

Recorded plan streaming retains the completed text, and no merge-blocking issue is established. A live plan-mode run remains unverified.

Pre-merge checks: 4

Passed checks: 4 passed

Check name	Status	Explanation
Title check		The title clearly identifies the main change: Codex plan streaming no longer commits every token. It uses a concise conventional-commit format.
Description check		The description explains the problem, fix, scope, measured impact, verification steps, and unverified live-run coverage. It references issue #18956 and provides focused test results. The headings diff…
Linked Issues check		Issue #18956 requires reducing repeated Codex item/plan/delta event commits while preserving plan completion behavior. The PR routes plan deltas through makeProviderTextDeltaCoalescer with a 400 m…
Out of Scope Changes check		The changes are limited to Codex plan-delta coalescing and its adapter test. The coalescer integration, finalization handling, and test coverage directly support issue #18956. No unrelated source, con…

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Usage-based review receipt

• Mode: Continue automatically
• Reviewed files: 2
• Charged: $0.50
• View usage details

Note

This review was completed with usage-based billing: files reviewed beyond your plan's included limits are billed at $0.25/file. View usage-based billing.

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

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	+7 B (+0.1%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	+7 B (+0.6%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: acb9027 · PR result: 82f35c8 · Source CI: success

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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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