add-chat-model

name	add-chat-model
description	Add a new chat model id to @aituber-onair/chat and wire it through constants, provider implementation, tests, example UI selector, docs, and versioning updates. Use when requests include adding a new model, supporting a specific model id, adding a provider model, or updating supported models for providers such as claude, gemini, and openai.

Add Chat Model

Overview

Add a new model id for an existing chat provider and complete all required code, test, docs, and version updates in one pass.

Before adding any model to a supported list, read docs/agent-model-provider-guidelines.md. A model is not eligible for getSupportedModels() unless the exact endpoint family used by the provider is documented or live-verified for that model.

Inputs

Collect missing inputs before editing:

• provider: one of claude, gemini, openai, or another existing provider
• model_id: model id string (example: provider-chat-model-id)
• model_const_name: exported const name (example: MODEL_PROVIDER_CHAT_MODEL)
• display_name: UI label (example: Provider Chat Model)
• supports_vision: boolean
• api_version_hint: optional hint for API version routing (example: v1beta only, v1 preferred, unknown)
• bump_version: boolean, default false; set true only when the user explicitly requests release/version work
• next_version: optional, explicit target version (example: 0.15.0)

Procedure

1. Run preflight checks before editing:
   • Read docs/agent-model-provider-guidelines.md and classify the candidate as recommended/default, supported/explicit, exported deprecated compatibility, or candidate-only.
   • Search for existing candidate notes or research notes: rg "<model_id>|<provider>|candidate|research|notes" docs skills .claude. Treat any matches as non-authoritative until official docs or live API behavior re-confirm them.
   • Grep existing usage to avoid duplicate work: rg "<model_id>" packages/chat.
   • Inspect provider constants and provider getSupportedModels() before adding anything.
   • If the model already exists, stop and switch to gap-fix mode (tests/docs/versioning only if missing).
2. Validate API family and user-facing eligibility before changing supported lists:
   • Confirm whether the model must use a specific API version or endpoint flavor (for example Gemini v1 vs v1beta).
   • Confirm the provider docs or live API behavior show this model working with the same endpoint family used by the current service implementation (for example Chat Completions vs Responses vs legacy completions).
   • Confirm request fields, response shape, streaming format, tool-call format, and vision input format match the current provider parser.
   • Confirm users can configure the model through package options and relevant example UI without undocumented setup.
   • Do not continue to step 3 when only a global model page confirms that the model exists.
   • If any hard gate fails, keep the candidate as candidate-only or export-only compatibility and do not add it to getSupportedModels() or example selectors.
   • Treat api_version_hint as a starting point, then verify in provider docs or behavior.
3. Locate the provider constants file and add:
   • export const <model_const_name> = '<model_id>';
   • Keep ordering consistent with neighboring entries.
   • For claude, edit packages/chat/src/constants/claude.ts.
4. If supports_vision is true, add the new constant to the provider vision list.
   • For claude, update CLAUDE_VISION_SUPPORTED_MODELS.
5. Update provider implementation so supported models include the new constant.
   • For claude, edit .

• Decide next version:
  • Use next_version if provided.
  • Otherwise use a minor bump for new model support unless repository rules differ.
• Update packages/chat/package.json version.
• Update packages/chat/CHANGELOG.md for the target version:
  • If a section for next_version already exists, append to it.
  • Do not create duplicate headers for the same version.
• Align dependent ranges:
  • Update packages/core/package.json dependency @aituber-onair/chat to the new range (for example ^0.14.0).
  • Do not bump @aituber-onair/core version only for this dependency range change.
• Update lockfiles affected by version/range changes:
  • package-lock.json
  • package/example lockfiles that embed workspace metadata when changed
• Follow repository rule: do not create .changeset/*.

11. Verify:

• Run lockfile/install sanity check.
• Run chat package tests.
• Run typecheck/build for chat package.
• Run core typecheck to ensure dependency range updates are valid.
• Grep for model_id to confirm expected placements and no duplicates.

12. Prepare final commit:

• Use a release-prep message such as: chore(release): prepare @aituber-onair/chat v<next_version>.
• Keep release-prep changes (version/changelog/lockfiles) in the same commit.

13. Offer core propagation handoff:

• After chat-side work is complete, ask: "Run $sync-core-after-chat-upgrade now to propagate this chat upgrade into @aituber-onair/core and core examples?"
• If the user agrees, execute that skill next with the same target chat_version.
• Skip the question only when the user already requested end-to-end chat+core propagation in the same task.

Verification Commands

Run commands from repository root:

npm ci
npm -w @aituber-onair/chat run test
npm -w @aituber-onair/chat run typecheck
npm -w @aituber-onair/chat run build
npm -w @aituber-onair/core run typecheck
rg "<model_id>" packages/chat
rg "v1beta|streamGenerateContent|generateContent" packages/chat/src/services/providers

Acceptance Criteria

• New model constant exists and is exported.
• Provider supported models contain the new model only when endpoint-family compatibility is documented or live-verified.
• If vision is enabled, vision support list contains the model and related tests pass.
• Example selector displays the model label.
• Service request routing matches model API requirements (version/endpoint).
• Service-level tests cover request path/version and fallback behavior when used.
• Final notes include the model decision record from docs/agent-model-provider-guidelines.md.
• English and Japanese README mention the model.
• Version and changelog are updated when bump_version is true.
• @aituber-onair/core dependency range is aligned to the new chat version.
• Lockfiles are consistent and npm ci succeeds.
• The user is asked whether to run $sync-core-after-chat-upgrade after chat updates finish (unless already requested explicitly).