You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(server): auto-resume OpenCode Go threads when usage limits reset - #31838
OpenCode 2 reports Go subscription limits as provider.go-limit, without a reset time. These failures previously remained failed instead of entering the existing automatic usage-limit recovery flow.
Change
Recognise Go limits and resolve their reset from the Go usage windows, waiting for the latest named or exhausted window. Perform the lookup outside the shared event lock, with a bounded timeout. Preserve Stop and replacement-start semantics so a late lookup cannot resume a stopped thread or end a replacement execution. HTTP 402 remains non-resettable.
Scope and approval
This is a focused repair of the existing opt-in automatic usage-limit recovery capability: OpenCode Go errors were missing the classification and reset timestamp that recovery already requires. It adds no setting or broader workflow. Submitted as an obvious bug fix without a prior approval discussion; no maintainer approval is claimed.
Verification
Validation: helper tests 8/8; adapter tests 104/106 (two pre-existing Windows attachment-path failures); provider suite 164 passed, 3 skipped, one pre-existing Windows XDG-path failure. Provider/server typechecks, lint and format pass. Gated regressions cover Stop during lookup, undecodable replacement starts and reconnect with a stale active listing. No live reset was exercised.
Review log (completed locally before publication):
Independent Opus rounds found shared-lock blocking, ambiguous window matching and a Stop race; fixed and final verdict happy.
Independent Sol rounds found undecodable replacement-start invalidation; fixed. Reconnect cancellation suggestion declined by Opus because a stale active listing could discard a genuine terminal failure; regression added. Final Sol verdict happy.
Trial GLM rounds flagged shared-lock blocking and HTTP 402; fixed. Final verdict happy, independently reproduced tests and checks.
Scope/limitations: OpenCode v1 has only a bare 429, insufficient to attribute a limit to Go, so remains unchanged. Unnamed legacy limits with lagging usage data may lack a reset time and require manual resume. Lost replacement-start events across reconnect remain an existing limitation. Lookup failure degrades to a usage limit without a reset. Commit email metadata was changed to GitHub no-reply after review; the final tree is identical to reviewed e24edba.
Implementation: Claude Opus 5.5 in the Claude Agent harness through T3 Code. Independent reviews: Opus 5.5, GPT-6.1 Sol (Codex harness), and GLM-5.3-Flash (OpenCode harness). Coordination and publication: GPT-6.1 Sol in the Codex harness through T3 Code.
OpenCode reports a spent Go window as `provider.go-limit` ("OpenCode Go 5 hour
usage limit reached…"). T3 classified that as a generic provider error with no
reset time, so limited threads never armed usage-limit recovery and stayed
failed after the window reset, unlike Claude and Codex threads.
The OpenCode 2 adapter now treats `provider.go-limit` (and the older
`provider.quota` 429 `GoUsageLimitError` shape) as a usage limit and takes its
reset time from OpenCode Go's usage windows, which the driver already reads
for the usage bars: the window the error names, or any other spent Go window
that resets later. When Go usage cannot be read the failure stays a usage limit
without a reset, as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Go limit fix read Go usage inside the adapter's event handler, which holds
the lock every OpenCode thread on the instance shares. Each Go limit failure
stalled all threads for up to the 5 second usage timeout, and an account-wide
limit hitting several threads serialized those reads.
A failed Go turn now keeps its end pending, reads usage in a forked fiber, and
ends under the lock once the read returns (or fails, or times out after 10
seconds, with no reset). A new execution, a Stop, or another end for that turn
supersedes the pending end; Stop ends it as interrupted instead of reporting
that nothing was running, and a reconnect leaves it to the pending read.
Review follow-ups:
- A Go limit error that names several windows, or names one alongside another
word the old patterns matched first ("weekly ... session"), now waits for
the latest named window instead of the first pattern in the list. "session"
is no longer read as the 5 hour window. The patterns assume OpenCode's
English error text.
- provider.go-limit with HTTP 402 stays a provider error, like provider.quota.
- Known limitation: a legacy provider.quota GoUsageLimitError whose message
names no window, while the usage endpoint still shows every window below
100%, keeps the usage limit without a reset rather than guessing one.
- OpenCode 1 (the v1 adapter) is out of scope: it reports limits as a bare
APIError 429 that does not say whose limit it is, so Go reset times would be
applied to other providers' rate limits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Stop sent while OpenCode Go's usage windows were read could lose to
the read, which then ended the turn failed with a reset time, so the
stopped thread resumed at the reset. The read now ends a stopped turn
as interrupted with no reset.
A start this build cannot decode now cancels the pending read, like a
decoded start, so the read cannot end the restarted execution.
A reconnect still lets a pending read finish the turn when OpenCode
lists the session as running: that execution's end was already
delivered, and the listing does not prove a new start.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Macroscope's review found this PR approvable — OpenCode Go limit failures now include reset times so the existing opt-in recovery feature can resume threads. Auto-resume remains off by default, and timed reads and Stop handling limit unintended continuation.
You can add or adjust custom eligibility rules. Learn more.
No actionable comments were generated in the recent review.
Walkthrough
The OpenCode v2 adapter now reads usage windows for recognized Go limit failures and uses available window data to set a reset time. The changes also update failure classification and handle interrupts, restarts, and reconnects while a usage read is pending.
Changes: OpenCode Go Limit Reset Handling
Layer / File(s)
Summary
Select Go limit reset times packages/provider-opencode/src/server/usageLimits.ts, packages/provider-opencode/src/server/usageLimits.test.ts
openCodeGoLimitResetAt selects the latest available reset time among windows named in the error and windows at 100% usage. Tests cover named windows, spent windows, lagging data, and missing reset times.
Wire the usage-limit reader packages/provider-opencode/src/server/driver.ts, apps/server/src/orchestration-v2/Adapters/OpenCode2AdapterV2.testkit.ts, packages/provider-opencode/src/server/adapter.ts
The driver passes its usage-limit effect to the v2 adapter. The replay runtime accepts and forwards an optional effect. A comment documents the OpenCode 1 behavior for 429 errors.
Handle Go limit failures and pending reads packages/provider-opencode/src/server/v2/adapter.ts, apps/server/src/orchestration-v2/Adapters/OpenCode2AdapterV2.test.ts
The adapter recognizes current and legacy Go limit errors, reads usage windows with a 10-second timeout, and applies a reset time when available. Interrupt, execution-restart, and reconnect handling account for a pending read. Tests cover classification, read outcomes, and lifecycle ordering.
sequenceDiagram
participant OpenCode
participant OpenCode2AdapterV2
participant readUsageLimits
participant openCodeGoLimitResetAt
OpenCode->>OpenCode2AdapterV2: Send Go limit failure
OpenCode2AdapterV2->>readUsageLimits: Read usage windows
readUsageLimits-->>OpenCode2AdapterV2: Return usage limits
OpenCode2AdapterV2->>openCodeGoLimitResetAt: Resolve reset time from error and limits
openCodeGoLimitResetAt-->>OpenCode2AdapterV2: Return reset time or null
OpenCode2AdapterV2-->>OpenCode: End turn with usage-limit failure
pingdotgg/t3code#33758: Both changes prevent delayed asynchronous work from terminating a replacement turn.
pingdotgg/t3code#31089: Both changes populate resetAt for OpenCode v2 usage-limit failures, using different response data.
Suggested labels:bugfix
Suggested reviewers:juliusmarminge
Merge Risk: Minimal · up to 4d2fd
OpenCode Go usage-limit failures now resolve a reset time so threads can auto-resume. Stop and replacement-start behavior is preserved. No concrete merge-blocking risk was found. A live Go reset has not been exercised, and unnamed legacy limits with lagging usage data may still need a manual resume.
Pre-merge checks: 4
Check name
Status
Explanation
Title check
The title clearly summarizes the main change: automatic recovery of OpenCode Go threads when usage limits reset.
Description check
The description includes the required Problem, Change, Scope and approval, and Verification sections. It explains the bug, the fix, why this focused repair needs no prior approval, and the validation …
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.
Full details: Description check
Explanation
The description includes the required Problem, Change, Scope and approval, and Verification sections. It explains the bug, the fix, why this focused repair needs no prior approval, and the validation results and limitations.
Finishing Touches
Generate unit tests (beta)
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
size:L100-499 changed lines (additions + deletions).vouch:unvouchedPR author is not yet trusted in the VOUCHED list.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
github-actions
Bot
added
vouch:unvouched
![macroscopeapp[bot]](https://avatars.githubusercontent.com/in/900172?s=60&v=4){kind=small}
Problem
OpenCode 2 reports Go subscription limits as
provider.go-limit, without a reset time. These failures previously remained failed instead of entering the existing automatic usage-limit recovery flow.Change
Recognise Go limits and resolve their reset from the Go usage windows, waiting for the latest named or exhausted window. Perform the lookup outside the shared event lock, with a bounded timeout. Preserve Stop and replacement-start semantics so a late lookup cannot resume a stopped thread or end a replacement execution. HTTP 402 remains non-resettable.
Scope and approval
This is a focused repair of the existing opt-in automatic usage-limit recovery capability: OpenCode Go errors were missing the classification and reset timestamp that recovery already requires. It adds no setting or broader workflow. Submitted as an obvious bug fix without a prior approval discussion; no maintainer approval is claimed.
Verification
Validation: helper tests 8/8; adapter tests 104/106 (two pre-existing Windows attachment-path failures); provider suite 164 passed, 3 skipped, one pre-existing Windows XDG-path failure. Provider/server typechecks, lint and format pass. Gated regressions cover Stop during lookup, undecodable replacement starts and reconnect with a stale active listing. No live reset was exercised.
Review log (completed locally before publication):
Scope/limitations: OpenCode v1 has only a bare 429, insufficient to attribute a limit to Go, so remains unchanged. Unnamed legacy limits with lagging usage data may lack a reset time and require manual resume. Lost replacement-start events across reconnect remain an existing limitation. Lookup failure degrades to a usage limit without a reset. Commit email metadata was changed to GitHub no-reply after review; the final tree is identical to reviewed e24edba.
Implementation: Claude Opus 5.5 in the Claude Agent harness through T3 Code. Independent reviews: Opus 5.5, GPT-6.1 Sol (Codex harness), and GLM-5.3-Flash (OpenCode harness). Coordination and publication: GPT-6.1 Sol in the Codex harness through T3 Code.