Question: Why is Planning Mode restricted to Claude models only?

[stefandevo]

Summary

Planning mode (lite/spec/full) is currently restricted to Claude models only in the UI. After investigating the codebase, I couldn't find any technical justification for this limitation.

Current Behavior

The restriction is implemented via isClaudeModel() checks in 3 UI files:

• apps/ui/src/components/views/board-view/dialogs/add-feature-dialog.tsx:183
• apps/ui/src/components/views/board-view/dialogs/edit-feature-dialog.tsx:123
• apps/ui/src/components/views/board-view/dialogs/mass-edit-dialog.tsx:239

When a non-Claude model is selected, the planning mode selector shows: "Planning mode requires Claude"

Investigation Findings

No Claude-Specific Code Found

1. Planning prompts are generic - The prompts in libs/prompts/src/defaults.ts contain standard instructions that any capable LLM should handle. No Claude-specific syntax or features are used.

2. Server-side has no restrictions - auto-mode-service.ts uses ProviderFactory.getProviderForModel() which routes to any provider. There's no isClaudeModel() check on the server for planning mode.

3. Simple marker system - Planning relies on detecting [SPEC_GENERATED] in the output, which any instruction-following model can produce.

4. Task format is standard - The task format (- [ ] T001: Description) is plain markdown that any model can generate.

Git History

The restriction was introduced in these commits:

1. 41b4869 (Jan 9, 2026) by @webdevcody - Added isClaudeModel import during "OpenCode model support" refactoring
2. e64a850 (Jan 10, 2026) by @Shironex - Added tooltips stating "Planning modes are only available for Claude Provider"

Neither commit message explains the technical reasoning for restricting planning to Claude.

Question

Why is planning mode restricted to Claude models?

Other providers like:

• Copilot (GPT-5.x, Gemini models)
• Cursor (various models)
• OpenCode (various models)
• Codex (GPT models)

...should theoretically be capable of following the planning prompts and generating structured specs.

Potential Impact

Users selecting non-Claude models (e.g., copilot-gpt-5.2-codex, cursor-auto) cannot use planning mode even though:

1. The server would execute it correctly
2. The prompts don't require Claude-specific capabilities
3. These models are capable of structured output

Suggested Action

Either:

1. Document the reasoning if there's a valid technical reason for this restriction
2. Remove/relax the restriction if it was added out of caution rather than necessity
3. Make it configurable so users can opt-in to planning mode for other providers

cc: @webdevcody @Shironex - Could you share the context behind this design decision?

[Shironex]

Plan mode had fundamental issues and bugs that are still not resolved or fixed. I would keep it limited to Claude models until we fix them, since allowing it to run on non-Claude models could cause more issues.

[Shironex]

I will close it to have all bugs that i want to fix before v0.14.0rc get merged moved it to #671