Effort selector dropdown hidden for GLM models — blocks access to GLM-5.2 High/Max effort tiers

Aperta Adatta ai principianti
  • Qualcosa non va?
#640 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
2/5
Tempo stimato
1-3 ore
Idoneità per principianti
68/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
typescript
Ambito
frontend

Direzione di ricerca

Inizia leggendo il gate effort-selector in src/components/chat/MessageInput.tsx e le voci delle capacità dei modelli in src/lib/provider-catalog.ts; poi individua dove sono definiti i modelli env-provider. Controlla come il menu a discesa gestisce supportedEffortLevels prima di scegliere quali modelli e livelli esporre. Il lavoro è completato quando gli utenti GLM possono selezionare i livelli di effort supportati e il valore di effort esistente continua a essere trasmesso nella richiesta di chat.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Summary

When using a GLM model (e.g. GLM-5.2 via Z.ai's Anthropic-compatible endpoint at https://api.z.ai/api/anthropic), the effort selector dropdown does not render in the chat composer. This prevents users from selecting GLM-5.2's High and Max reasoning effort tiers, even though the downstream plumbing already carries the effort field end-to-end and Z.ai's endpoint accepts it.

Environment

  • CodePilot: v0.56.0 (source HEAD b4e96a1)
  • OS: Windows 11
  • Provider: env (Claude Code SDK runtime)
  • Endpoint: https://api.z.ai/api/anthropic
  • Model: glm-5.2[1m]

Root cause

The dropdown visibility is gated by a model capability flag that GLM does not have.

src/components/chat/MessageInput.tsx:991-992:
```tsx
const currentModelMeta = currentModelOption as (typeof currentModelOption & { supportsEffort?: boolean; supportedEffortLevels?: string[] }) | undefined;
const showEffortSelector = currentModelMeta?.supportsEffort === true;
```

supportsEffort: true is set on Claude variants in src/lib/provider-catalog.ts (Sonnet / Opus / Haiku roles, plus Fable 5 at lines 240, 250, 260, 290, 301, 314, 325, 344, 355, 370, 385, 396, 428, 437) but not on GLM models configured via the env provider. When a user selects GLM, currentModelMeta.supportsEffort is undefined, so showEffortSelector is false and the dropdown never renders. There is no way inside the UI to set selectedEffort to anything other than the 'auto' default.

The downstream plumbing already works

Once selectedEffort is anything other than 'auto', the value flows through cleanly:

src/app/chat/page.tsx:890:
```tsx
...(selectedEffort && selectedEffort !== 'auto' ? { effort: selectedEffort } : {}),
```

src/lib/claude-model-options.ts:102-144 (sanitizeClaudeModelOptions) passes effort through unchanged for any model that isn't Opus 4.7/4.8/Fable 5. GLM matches neither the OPUS_ADAPTIVE_THINKING_PATTERN nor the FABLE_PATTERN, so effort: 'max' would reach Z.ai's endpoint as-is.

The EffortSelectorDropdown itself also already falls back to the full 5-level set when supportedEffortLevels is undefined:

src/components/chat/EffortSelectorDropdown.tsx:36:
```tsx
const baseLevels = supportedEffortLevels || ['low', 'medium', 'high', 'xhigh', 'max'];
```

So the only thing missing is the capability flag on GLM model entries.

GLM-5.2 does support effort tiers

Per Z.ai's "How to Switch Models" doc:

"In a Claude Code session, type the /effort command to switch thinking intensity. After switching, Claude Code maps the selected effort to GLM-5.2's actual effort levels..."

GLM-5.2 exposes at least High and Max effort tiers. Users running the Claude Code CLI directly against Z.ai can reach them via /effort max. CodePilot users cannot, because the UI gate is closed. (CodePilot also does not forward /effort as a slash command to the underlying SDK, so there's no escape hatch.)

Suggested fixes

In order of preference:

  1. Mark env-provider models with supportsEffort: true by default so the dropdown renders for any Anthropic-compatible endpoint (GLM, DeepSeek, Kimi, etc.). Downstream code handles unknown effort values gracefully, and the endpoint will either accept the level or ignore it.

  2. Add explicit GLM entries to ENV_CLAUDE_CODE_MODELS with supportedEffortLevels: ['high', 'max'] matching what Z.ai's endpoint actually documents, so the dropdown surfaces only the levels GLM-5.2 truly supports.

Workaround

None inside CodePilot today. Switching to Opus surfaces the dropdown (Opus has supportsEffort: true), but that's not a real fix for users who chose GLM for cost or quota reasons.

Lingua principale
TypeScript
Stelle
6.5k
Fork
697
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

  • Nessun Dockerfile né file Docker Compose
  • Ha un modello di pull request
  • Nessuna guida per i contributori

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di op7418/CodePilot

Tutte le issue di op7418/CodePilot

Issue simili

Altre issue su TypeScript

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.