Source model list from /provider/v1/models API instead of scraping CLI bundle

[KRoperUK]

Problem

The model list in models.json is generated by scripts/sync-models.ts, which downloads the command-code npm tarball and parses the minified JS bundle using fragile string anchors (e.g. 'SONNET_4_6:{id:"claude-sonnet-4-6"'). This drifts out of date as the bundle changes: models.json currently has 21 models, while Command Code now serves 30.

Missing models include xiaomi/mimo-v2.5-pro, xiaomi/mimo-v2.5, claude-opus-4-8, claude-fable-5, moonshotai/Kimi-K2.7-Code, MiniMaxAI/MiniMax-M3, Qwen/Qwen3.7-Plus, stepfun/Step-3.7-Flash, and nvidia/nemotron-3-ultra-550b-a55b.

Change

Use the public, OpenAI-compatible endpoint GET https://api.commandcode.ai/provider/v1/models as the source of truth for which models exist, their display names, and context windows.

• The CLI bundle is still downloaded, but only to enrich entries with pricing (the endpoint does not expose cost).
• Fields the endpoint lacks (reasoning, tool_call) come from a small MODEL_META overrides map with sensible defaults (reasoning: true, tool_call: true).
• Tier is derived from the id namespace (namespaced → open-source, bare → premium), replacing the provider-based TIER_MAP which can't work since the endpoint reports owned_by: "command-code" for everything.
• Models without bundle pricing are kept with zeroed cost and a warning, rather than being silently dropped — so newly released models appear immediately. (7 models currently fall into this bucket; FALLBACK_COSTS can be populated as pricing becomes known.)

Removed

Brittle catalog scraping: extractModelCatalog, extractSpecConstants, HARDCODED_EXTRAS, and the provider-based TIER_MAP.

Artifacts

• Regenerated models.json (21 → 30 models) and the README table.
• Added tests/unit/models-json.test.ts to guard the models.json shape (id/name/tier/cost/limit, unique ids).

Verification

• bun run sync lists 30 models including xiaomi/mimo-v2.5-pro (1M context); no anchor/parse exceptions.
• bun run generate-readme updates the table.
• bun test tests/unit/ → 76 pass, 0 fail.

Notes / open questions

• This still relies on the bundle for pricing. A natural follow-up would be moving all cost data to a maintained FALLBACK_COSTS/overrides file and dropping bundle parsing entirely. Happy to do that here or separately — flagging since it changes the maintenance model.

🤖 Generated with Claude Code

[KRoperUK]

Reopened as #6 — rebased to stack cleanly on #5/#1. (GitHub wouldn't reopen this one after the branch history was rewritten.)