Status

The Mattermost model picker is LIVE production code on HEAD (2026-04-19) and continues working despite the deletions that #2337 predicted would force its removal.

Evidence that #2337's removal premise is falsified

#2337 listed the picker under "Discord/Mattermost model-picker UI cascade (commands-models.ts deletion)" — the stated rationale was that the picker depended on handleModelsCommand / buildModelsProviderData from the Pi-era commands-models.ts stub, so gutting the stub meant gutting the picker.

What actually happened:

• commands-models.ts was deleted in gut: eliminate callers and delete ~14 EXCLUDE-STUB files in src/auto-reply/ #2373 (merged 2026-04-16).
• The Mattermost picker has been stable since. It sources provider data from its own path, not from handleModelsCommand / buildModelsProviderData.
• git grep -n "handleModelsCommand\|buildModelsProviderData" extensions/mattermost/ returns zero hits — the picker never actually depended on those symbols.

Conclusion: the coupling claim in #2337 was wrong. The picker's inclusion in the cascade was based on a premise that doesn't hold. Removing a working user-facing feature on a falsified premise is not justified.

Default: Keep + verify

Absent a fresh product rationale, the picker stays. It's a working user-facing feature serving Mattermost channels.

Files still live (reference)

AC under Keep (default path)

• End-to-end verification: issue /model in Mattermost, confirm picker renders, confirm selection applies.
• Document that picker is preserved intentionally (not in gut: re-apply provider/model cascade after v2026.3.7 sync regression (~19 files) #2337's cascade scope) — either in extensions/mattermost/README.md or as an inline note where relevant.
• pnpm check + pnpm test green.

Alternative: Gut (requires fresh product rationale)

The original #2337 rationale (stub coupling) is dead. Gutting now requires a standalone product case such as:

• Model-selection UI is architecturally out of scope for RemoteClaw (e.g., runtime config owns model selection, not per-channel UI).
• Maintenance cost of Mattermost-specific picker exceeds its value.
• Strategic direction removes per-channel model switching entirely.

If product decides to gut on fresh grounds, the cascade is ~700 LoC:

• extensions/mattermost/src/mattermost/model-picker.{ts,test.ts} — 515 LoC
• Picker handling in monitor.ts (lines 56, 495, 778-887) — ~150 LoC
• Picker resolution in slash-http.ts (lines 30-32, 387) — ~5 LoC
• Any picker re-exports in src/plugin-sdk/mattermost.ts

AC under Gut (if fresh rationale provided)

• Fresh product rationale documented in issue comment, signed off by product owner.
• Files deleted + cascade stripped from monitor.ts / slash-http.ts / plugin-sdk.
• pnpm check / pnpm test green.
• Verification: git grep -rn "model-picker\|ModelPicker" extensions/mattermost/ src/plugin-sdk/mattermost.ts returns zero.

Relation to #2337

#2337's cascade scoped this picker for deletion based on a coupling claim that was falsified by #2373's real-world execution. Resolution in either direction — Keep (with verification + documentation) or Gut (with fresh rationale) — unblocks closing #2337. Part of #2337.