[PF-15] Doctor first screen simplified; doctor ai uses planforge.json and planforge model flow

[chucoding]

Purpose

• When running planforge doctor, show only mode selection (static / ai) on the first screen; remove the initial Doctor AI config (model list) block.
• Make doctor ai use planforge.json only, and use the same model selection flow and catalog (models.json) as planforge model when changing models.

Description

Related Notion: doctor

1. Doctor entry (first screen)

Before	After
Show Doctor AI config (default), then mode selection	Show mode selection (static / ai) only

• cli-js: In runDoctorModeSelect(), remove the block that calls getDefaultDoctorAiConfig and printCurrentAiConfig.
• cli-py: In run_doctor_mode_select(), remove the same block.

2. Doctor ai config source

Before	After
templates/doctor default-*.json (getDefaultDoctorAiConfig)	planforge.json (loadConfig)

• Initial planner/implementer and the config shown in the UI come from planforge.json.
• UI title: "Doctor AI config (planforge.json)".
• cli-js and cli-py: doctor ai no longer call getDefaultDoctorAiConfig / get_default_doctor_ai_config.

3. Doctor ai model selection = planforge model flow

• Catalog (models.json) required: If loadModelsCatalog() fails, doctor ai exits with an error message.
• Change models: After choosing "Role to change" (planner or implementer), call runModelTui (cli-js) / _run_model_tui (cli-py) twice with preselectedMode / preselected_mode. Provider → Model → Effort/Reasoning selection is the same as planforge model.
• Non-interactive: --provider / --model are only accepted if the pair exists in the catalog.

Files changed

Area	File	Summary
cli-js	src/commands/doctor.ts	First-screen config removed; planforge.json used; runModelTui used for Change models; catalog required; removed buildOptionsFromCatalog, selectProviderAndModel, etc.
cli-js	src/commands/model.ts	Added options?: { preselectedMode?: string } to runModelTui
cli-js	src/config/load.ts	Removed getDefaultDoctorAiConfig import from doctor (function kept for other use)
cli-py	planforge/commands/doctor.py	First-screen config removed; planforge.json only; _run_model_tui for Change models; catalog required; Run immediately / Change models loop
cli-py	planforge/commands/model.py	Added preselected_mode to _run_model_tui

How to test

1. First screen: Run planforge doctor in a TTY. Confirm that only the mode list (static / ai / Quit) is shown, with no Doctor AI config block at the top.
2. Doctor ai config: Run planforge doctor → choose ai (or planforge doctor ai). Confirm "Doctor AI config (planforge.json)" and that planner/implementer match planforge.json.
3. Change models: Choose "Change models" → choose planner or implementer for "Role to change" → confirm the same Provider → Model → Effort/Reasoning flow as planforge model runs twice, then the Action menu appears again.
4. Catalog required: Temporarily move or rename packages/core/models.json, then run planforge doctor ai. Confirm it exits with a message like "doctor ai uses the same model catalog as planforge model...".
5. Non-interactive: Run planforge doctor ai --provider claude --model claude-haiku-4-5 and confirm tests run. Run with a provider/model not in the catalog and confirm it fails with a "Check models.json catalog" (or similar) message.
6. cli-py: Repeat the above with the Python CLI where applicable and confirm behavior matches.

Review Requirement

• Confirm doctor ai only uses planforge.json and that getDefaultDoctorAiConfig / get_default_doctor_ai_config are not used on the doctor/doctor ai code paths.
• For runModelTui / _run_model_tui with preselectedMode(preselected_mode), confirm planner and implementer are each selected once and the result is applied correctly.
• Spot-check that Node and Python behavior match (first screen, config source, Change models flow, catalog requirement, non-interactive validation).

Additional Info

• Related Notion: doctor — updated via MCP to reflect this change (first screen, planforge.json, model flow integration).
• Aligned with .cursor/rules/node-python-symmetry.mdc: both cli-js and cli-py updated.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.