OpenClaw Kimi K2.6 Bugfix

The Problem

A common OpenClaw issue is Kimi configuration drift.

The same installation can end up with conflicting Kimi settings across the global config, agent-local config files, auth profiles, and saved session state.

When that happens, OpenClaw often fails over to a fallback model without making the root problem obvious — and that fallback can then persist across future turns.

The Fix

Normalize the entire Kimi path end-to-end:

• Keep only the intended moonshot/kimi-k2.6 configuration
• Point it to https://api.kimi.com/coding
• Remove legacy kimi / kimi-coding entries and older Kimi variants
• Clear any stale fallback state so sessions actually stay on K2.6

Prompt

please fix my OpenClaw Kimi config. i have the bug where sessions keep falling back to another model instead of staying on moonshot/kimi-k2.6. this issue often happens because Kimi settings drift across both global and agent-local config files, and stale fallback state can keep the wrong model active even after partial fixes.

please do this carefully and verify exhaustively:

1. inspect global /root/.openclaw/openclaw.json
2. inspect every agent-local file under ~/.openclaw/agents/*/agent/, especially:
   - models.json
   - auth-profiles.json
   - auth-state.json
3. remove all legacy kimi and kimi-coding provider/auth entries everywhere, not just the first block you find
4. keep only moonshot/kimi-k2.6
5. make sure its base URL is exactly https://api.kimi.com/coding
6. verify there are no leftover Kimi variants anywhere, including:
   - k2p5
   - kimi-code
   - kimi-k2.5
   - kimi-k2-thinking
   - kimi-k2-thinking-turbo
   - kimi-k2-turbo
7. clear stale session/provider fallback state that pins sessions to the fallback model
8. reload/restart if needed
9. verify the active session actually runs on Kimi after the cleanup

important: don't do a shallow check. after editing, explicitly confirm every relevant file only contains the intended moonshot provider and kimi-k2.6 model.