fix(models): omit rejected GLM Flash effort on OpenCode Go - #102
AojdevStudio wants to merge 2 commits into
Conversation
Bugbot needs on-demand usage enabledBugbot uses usage-based billing for this team and requires on-demand usage to be enabled. A team admin can enable on-demand usage in the Cursor dashboard. |
There was a problem hiding this comment.
ℹ️ No critical issues. Two small suggestions, one inline and one below.
Reviewed changes
Reviewed the whole diff. It stops the direct OpenCode Go GLM Flash route from sending an effort variant, and it keeps the OpenRouter ladder.
- Direct route sends no variant:
effort: []makesgetModelEffortLevelsreturn an empty ladder.resolveRungthen returnsundefined, sorunPromptTurnleaves outvariantfor this route.openRouterEffortis now explicit, so the OpenRouter route andossEffortFloorstill getlow, high, max. - Regression test: the new
test/models.test.tscase can fail. The old ladder resolved position0.5tohighon the direct route, and the test now requiresundefined. - Catalog drift exception: the ladder test in
models-catalog.main.test.tsnow skipsopencode-go/glm-flashand only checks thateffortequals[].
ℹ️ Nitpicks
test/models-catalog.main.test.ts:167— the special case turns off the drift check for this route entirely. If models.dev stops listing the ladder, nothing will say the exception is no longer needed. Addingexpect(rungs(publishedEffort(alias.resolve))).toBeDefined()before thereturnwould make the test fail at that point, which prompts someone to remove the exception.
Claude Opus (free via Pullfrog for OSS) | 𝕏
|
@pullfrog review the current head after the follow-up that addressed your inline comment. Do not make code changes. |
There was a problem hiding this comment.
✅ No new issues found. Both points from the last review are fixed.
Reviewed changes
Reviewed cdaa458, which changes only a code comment and one test assertion since the last Pullfrog review.
- Rewrote the
glm-flashcode comment: the comment atmodels.ts:877-878now quotes the provider's error text. The cross-repo#964and run-ID reference is gone, so the comment no longer points at a pullfrog issue number that has nothing to do with this. - Added a check that the drift exception is still needed:
test/models-catalog.main.test.ts:168now assertsrungs(publishedEffort(alias.resolve))is defined.rungsreturnsundefinedwhen the list is empty or missing, so this fails if models.dev stops listing any effort rung foropencode-go/glm-5.3-flash. That failure tells someone to remove the special case.
Claude Opus (free via Pullfrog for OSS) | 𝕏
|
Coordination check: head cdaa458 has the clean Pullfrog review at #102 (review), success review check, and no unresolved inline threads. The exact-head Test workflow is action_required: https://github.com/pullfrog/pullfrog/actions/runs/36474224896. A maintainer must approve that fork workflow run, then its tests must pass. I have made no additional code changes, billing changes or gate bypasses, and have not merged or enabled auto-merge. |

Problem
Pullfrog's homelab review run 36464251799 used OpenCode Go's
glm-5.3-flashat default high effort. The agent completed several tool calls, then the provider rejected a request withnative reasoning control reasoning_effort is not allowed. Pullfrog's model catalog advertises an effort ladder for this route, so its OpenCode adapter sends the selected variant. The live provider behavior disagrees with the published models.dev metadata.Change
Omit effort variants for the direct OpenCode Go GLM Flash route. Keep the existing GLM Flash model and provider. Preserve the separate OpenRouter effort ladder for runs already configured to use that route. Add a regression test and record the observed catalog exception.
Evidence
highfor the direct route.bunx vitest run test/models.test.ts effort.test.ts: 368 passed.bun run typecheck: passed.GPT-6-Sol in the Codex harness through T3 Code made these changes.