Telegram: selecting native Codex Ultra in /models is persisted as Max

[violetaini]

Version

Pikiloom v0.4.71 / ghcr.io/xiaotonng/pikiloom:0.4.71 (verified in the published ARM64 image).

Summary

The GPT-5.6 adaptation is working in the model catalog, dashboard dispatch, and Codex driver. There appears to be one remaining Telegram-specific path that still converts native Codex ultra to max.

The Telegram /models confirmation path calls bot.switchEffortForChat(...):

• command-ui.ts
• command-ui.ts

switchEffortForChat() currently applies the old agent-agnostic conversion:

const ultra = effort === 'ultra';
const realEffort = ultra ? 'max' : effort;

Source: src/bot/bot.ts

For Codex, workflow capability is false, so selecting Ultra through Telegram stores max, the picker subsequently reports max, and the app-server receives max rather than native ultra.

The dashboard path already uses splitEffortForAgent(), which correctly returns:

splitEffortForAgent('codex', 'ultra')
// { effort: 'ultra', workflow: false }

Reproduction

1. Select the codex agent and gpt-5.6-sol (or gpt-5.6-terra).
2. In Telegram, open /models.
3. Select Ultra and confirm.
4. Reopen /models or inspect the next Codex turn.

The effective effort is max instead of ultra.

Expected behavior

• Codex max remains max.
• Codex ultra remains ultra and is sent verbatim to the app-server.
• Claude keeps its existing display alias behavior: ultra -> max + workflow orchestration.

Suggested fix

Reuse splitEffortForAgent(cs.agent, effort) inside switchEffortForChat() instead of performing the unconditional ultra -> max conversion, and add a Telegram /models regression test for Codex Ultra.