fix(server): a Claude usage limit no longer resets the context meter to 0%

[Vantrongs]

Problem

When a Claude turn hits the subscription usage limit, the context meter drops to "0% · 0/1m" and stays there until the next successful turn. On a 429, Claude Code emits a locally generated assistant frame (model: "<synthetic>", error: "rate_limit") whose usage is all zeros. The root-assistant branch in ClaudeAdapterV2 turns every root message.usage into the provider turn's tokenUsage, so the failed turn reports usedTokens: 0. latestNativeContextUsage then picks that newest turn, and the provider thread's contextUsage becomes { usedTokens: 0, maxTokens: 1000000 }.

The zero also feeds shouldOfferResumeCompaction. The offer to compact before resuming a Claude thread that has been idle for 70+ minutes with 100k+ tokens doesn't appear for a thread stopped by the 5-hour limit, which is the case it exists for.

Seen on 0.0.46-nightly.20261005.2676 (Linux desktop, Claude Code 2.1.289): the last completed turn stored usedTokens: 244541, and each of the six usage-limited turns after it stored 0.

Before: the thread after its usage-limited turns, with the context meter at "0% · 0/1m":

Change

The adapter no longer reports a <synthetic> frame's usage as the turn's context. These frames are generated by the CLI and never reach the model. The adapter already handles them separately for goals, and usagePricing.ts excludes them as never billed. Sign-in errors and /goal status come as the same frames, so they no longer zero the meter either. Billed turn usage (turnTokenUsage, from the result frame) is unchanged.

Scope and approval

A small, focused fix of an obvious bug: one condition in one adapter, with no contract or client change.

Verification

• New test keeps the measured context when a usage limit ends the turn replays a real assistant frame, the CLI's synthetic rate_limit frame and a 429 result. Before the fix the provider turn reported the context as [2, 0]; with it, [2].
• vp test run apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts: 144 passed.
• tsc --noEmit -p apps/server/tsconfig.json: no errors. vp fmt --check and vp lint on the two files: clean, apart from the existing unused layer warning.
• Not checked in a running app: that the meter keeps its value through a real usage limit.

Model and harness: Claude Opus 5.5, Claude Code in T3 Code

[macroscopeapp]

Approvability

Verdict: Approved at 62a4038

Macroscope's review found this PR approvable — This is a narrowly scoped server bug fix that filters invalid zero-usage synthetic Claude frames while preserving real usage reporting. It includes a focused regression test and does not change schemas, defaults, integrations, or sensitive functionality.

^{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: Advanced
• Run ID: d2f82ac7-631d-4875-9f1c-e97879d4bf8a

---

📥 Commits

Reviewing files that changed from the base of the PR and between a825d2e and 62a4038.

---

📒 Files selected for processing (2)

• apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts
• apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts

---

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 6 remain after this review.

---

---

📝 Walkthrough

Walkthrough

The Claude adapter no longer updates token usage from synthetic root assistant frames. A regression test checks that a failed rate-limited turn retains its previously measured context usage.

Changes

Claude usage reporting

Layer / File(s)	Summary
Filter synthetic usage frames apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts, apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts	The adapter excludes root assistant frames with model "<synthetic>" from usage updates. A regression test checks that a failed 429 turn retains its measured context usage.

Priority: ⬇️ Low

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

Change: Bug fix

Suggested reviewers: juliusmarminge

---

Merge Risk: ⚪ Minimal · up to 62a40

No actionable issue remains from the supplied evidence; the change is mergeable after normal checks.

Architecture Summary

Architecture risk: 🔵 Low · up to 62a40

The change affects 1 system.

Changed systems: apps/server

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — apps/server (service) was modified; 2 changed files map to changed impact.

Before / after behavior

• observed — Modified behavior in apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts: Adds a runtime test that sends assistant text followed by a synthetic zero-usage rate_limit frame and a terminal 429 API error, then verifies the failed turn still reports measured context usage of 2 tokens.
• observed — Modified behavior in apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts: The root-assistant usage update now excludes "<synthetic>" model frames; previously any root assistant frame with usage data updated token usage.

---

Pre-merge checks | 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main fix: preventing Claude usage-limit errors from resetting the server context meter to zero.
Description check		The description covers the problem, root cause, change, scope rationale, focused regression test, validation commands, observed results, and the unverified running-app check. It is complete and aligne…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

---

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

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

---

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

Dismissing prior approval to re-evaluate a825d2e

Dismissing prior approval to re-evaluate 62a4038