Repository navigation
Conversation
…Org#2055) OpenCode Go Responses-format models now send temperature only for verified model and reasoning-effort combinations. This applies to both the streaming and non-streaming Responses requests. The settings UI uses the same rule: it hides the temperature control when the value would not be sent. Hiding the control does not clear or reset a saved temperature. - grok-4.6, muse-spark-1.3-contributor and muse-spark-1.2-contributor: temperature is sent at any reasoning effort. - gpt-5.6-luna and gpt-6-luna: temperature is sent only when the effective reasoning effort is "none". At other efforts the gateway returns HTTP 400 for every value except 1, which is the OpenAI reasoning default rather than real support, so the value is left out. - Temperature is sent only when the user has set a custom value. The provider default is never sent on this route. Changes: - types: add OPENCODE_GO_TEMPERATURE_REASONING_EFFORTS and opencodeGoResponsesSupportsTemperature(), and set supportsTemperature on every Responses model. - handler: add getResponsesTemperature(), which reads the raw modelTemperature setting and applies the opt-in rule; remove the unused temperature parameter from streamResponsesMessage; update the route comments so they no longer say the Responses route is Luna-only. - webview: add isTemperatureControlSupported(), which ApiOptions uses to decide whether to show TemperatureControl. It works out the effective reasoning effort the same way the request builder does. - tests: cover the rule, the model registry, the fetcher merge, both request modes, a stale "none" effort while reasoning is switched off, and the UI visibility round trip with the saved value kept. Shared transforms (getModelParams, RouterProvider) are unchanged. Refs Zoo-Code-Org#2006 Closes Zoo-Code-Org#2055 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📜 Recent review details
📝 Summary
Merge Risk: ⚪ Minimal · up to
|
| Check name | Status | Explanation |
|---|---|---|
| Title check |  |
The title "feat(opencode-go): opt-in temperature for Responses models (#2055)" matches the main change: OpenCode Go Responses-format models now send temperature only for verified model and reasoning-e… |
| Description check | |
The description follows the repository template and is mostly complete. It explains the behavior, the per-model rules, the file-level implementation, and the test procedure. It links the related issue… |
| Linked Issues check | |
The changes satisfy the coding requirements in #2055. The OpenCode Go capability helper uses an opt-in model and effective-reasoning-effort rule. getResponsesTemperature() applies the rule in both s… |
| Out of Scope Changes check | |
The changed source files, tests, visual fixture, and gallery story support the #2055 request. They implement or verify OpenCode Go Responses temperature capability, request behavior, model metadata, a… |
| Regression Evidence | |
Focused regression coverage is present. The request tests cover streaming and non-streaming forwarding, supported and unsupported reasoning efforts, stale disabled reasoning, unset/null values, and ex… |
| Security Boundaries | |
No changed security-boundary violation is present. The new request path adds temperature only after typeof temperature === "number" and opencodeGoResponsesSupportsTemperature() accepts the exact… |
| Persistence Integrity | |
No changed persistence path meets the failure conditions. The PR changes request filtering and UI visibility only. getResponsesTemperature() reads the saved modelTemperature and omits it for unsup… |
| Lifecycle Resource Cleanup | |
No changed lifecycle path introduces a resource leak or duplicate work. The PR adds synchronous temperature capability checks and request-body fields. The Responses streaming path still passes the exi… |
Full details: Title check
-
Explanation
The title "feat(opencode-go): opt-in temperature for Responses models (
#2055)" matches the main change: OpenCode Go Responses-format models now send temperature only for verified model and reasoning-effort combinations. It names the provider, the feature, and the scope. The issue number suffix is harmless, and the title is a single, specific sentence.
Full details: Description check
-
Explanation
The description follows the repository template and is mostly complete. It explains the behavior, the per-model rules, the file-level implementation, and the test procedure. It links the related issues (Refs
#2006, Closes#2055) and fills in the checklist and the visual snapshot N/A rationale. Minor gaps: the Test Procedure cites grok-4.7, which the rules do not cover, and the 'Get in Touch' and 'Videos' sections are filled in. The Description section uses some terms such as 'handler' and 'webview' without explanation. The author's own description also includes some non-Simplified Technical English phrasing, such as the parenthetical notes and the 'Observed varying degrees of performance' sentence, but this does not make the description incomplete.
Full details: Linked Issues check
-
Explanation
The changes satisfy the coding requirements in
#2055. The OpenCode Go capability helper uses an opt-in model and effective-reasoning-effort rule.getResponsesTemperature()applies the rule in both streaming and non-streaming Responses requests and omits unset or unsupported values. The UI uses the same rule and preserves saved temperature settings. Tests cover model metadata, fetcher merging, request forwarding and omission, reasoning changes, and UI visibility. The existing 0–2 temperature control and its resolution remain unchanged.
Full details: Out of Scope Changes check
-
Explanation
The changed source files, tests, visual fixture, and gallery story support the
#2055request. They implement or verify OpenCode Go Responses temperature capability, request behavior, model metadata, and matching UI visibility. No unrelated product behavior or unrelated refactor is demonstrated.
Full details: Regression Evidence
-
Explanation
Focused regression coverage is present. The request tests cover streaming and non-streaming forwarding, supported and unsupported reasoning efforts, stale disabled reasoning, unset/null values, and explicit temperature 0. Type and UI helper tests cover registry capabilities, unknown/non-Responses models, model-info loading, provider boundaries, and effective effort changes. The fetcher merge has focused coverage. The durable settings change now has a Playwright component test, gallery story, and committed
opencode-go-temperature.pngbaseline with supported and hidden states. Prior failure addressed: The PR now adds OpencodeGoTemperature.visual.tsx, its gallery story and fixture, plus the committed screenshots/opencode-go-temperature.png baseline. Existing ApiOptions Vitest coverage remains.
Full details: Security Boundaries
-
Explanation
No changed security-boundary violation is present. The new request path adds
temperatureonly aftertypeof temperature === "number"andopencodeGoResponsesSupportsTemperature()accepts the exact Responses-model allowlist and approved effort mapping. Unknown and non-Responses model IDs return false. The UI uses the same allowlist rule. The changes do not expose secrets or PII, execute input, or bypass approval controls.
Full details: Persistence Integrity
-
Explanation
No changed persistence path meets the failure conditions. The PR changes request filtering and UI visibility only.
getResponsesTemperature()reads the savedmodelTemperatureand omits it for unsupported configurations without modifying the stored value.ApiOptionspasses the existing configuration through, and the added test verifies that switching reasoning effort causes nomodelTemperaturewrite while restoring the saved value when support returns. No new unawaited write, non-atomic persistence operation, rollback gap, or unpropagated persisted default is introduced.
Full details: Lifecycle Resource Cleanup
-
Explanation
No changed lifecycle path introduces a resource leak or duplicate work. The PR adds synchronous temperature capability checks and request-body fields. The Responses streaming path still passes the existing abort signal and closes its captured async iterator in
finallyviastreamIterator.return?.()(src/api/providers/opencode-go.ts:431-491). The new UI helper and gallery fixture add no listeners, watchers, timers, providers, or tasks.
✨ Finishing Touches
-
🧪 Generate unit tests (beta)
-
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Comment @coderabbitai help to get the list of available commands.
Review statusThanks for contributing. This comment tracks the review sequence and the next action. Current step: The required review sequence passed. Remaining merge requirements apply. Review-state labels are managed by this workflow; do not edit them manually. |
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
github-actions
Bot
added
coderabbit-review-active
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
github-actions
Bot
added
awaiting-maintainer
github-actions
Bot
added
coderabbit-review-active
github-actions
Bot
added
awaiting-maintainer
taltas
left a comment
taltas
left a comment
There was a problem hiding this comment.
Looks clean to me, thanks for the contribution 🙌
github-actions
Bot
removed
the
awaiting-maintainer
OpenCode Go Responses-format models now send temperature only for verified model and reasoning-effort combinations. This applies to both the streaming and non-streaming Responses requests. The settings UI uses the same rule: it hides the temperature control when the value would not be sent. Hiding the control does not clear or reset a saved temperature.
Related GitHub Issue
Refs #2006
Closes #2055
Description
Shared transforms (getModelParams, RouterProvider) are unchanged.
Test Procedure
Tested extensively with all 5 named models, and tested with the prompt enhancement feature. Observed varying degrees of performance and generally functional. (noticed poor performance and high failure rate with the extreme low and extreme high temperature values - seems typical of the model responses observed while testing #2006).
Pre-Submission Checklist
*.visual.tsxsnapshot inwebview-ui/. Seewebview-ui/AGENTS.md→ "When a UI change needs a snapshot".Visual Snapshots
N/A. The change is behavior-only: it changes when the existing temperature control is shown for OpenCode Go Responses models, and the control's look is unchanged. The visibility rule is covered in Vitest (temperatureSupport.spec.ts, ApiOptions.spec.tsx), per webview-ui/AGENTS.md.
Videos (interaction / animation only)
https://github.com/user-attachments/assets/831a9fbb-d22b-45ca-98ef-85b0f97d597e
Shows some testing of various models using streaming path (via Ask agent) + mitmproxy to verify request payload
https://github.com/user-attachments/assets/1d7d4d5a-3028-4753-901d-7107170b1ea8
Shows some testing of various models using non-streaming path (via prompt-enhancement tab) + mitmproxy to verify request payload
Documentation Updates
[*] No documentation updates are required.
No docs page in this repo explains OpenCode Go settings or temperature. OpenCode Go appears only in the README's release highlights and in CHANGELOG/README translations, which are updated in bulk at release time.
The change is self-explanatory in the UI: the temperature control appears when it works and is hidden when it doesn't, and a saved value is never lost.
The rationale (Luna accepts temperature only at reasoning effort "none") goes in the [ENHANCEMENT] Support verified temperature capability for OpenCode Go Responses models #2055 closing comment and the code JSDoc, not user docs.
Get in Touch
Discord: anthony_25019