v1.30.2

Documentation

ccmm now defaults to MiniMax M3.1 Flash (#238)

The MiniMax recipe in the Alternative LLM
Providers
guide was pinned to MiniMax-M3[1m] and set auto-compaction to the
model's full one-million-token context window. Both are now corrected:

• The main, Sonnet, Opus, and Haiku slots default to
  MiniMax-M3.1-Flash[1m], matching MiniMax's current Claude Code
  guide. Flash always enables thinking.
• CLAUDE_CODE_AUTO_COMPACT_WINDOW drops to 524288 — the window
  MiniMax actually recommends. Compacting only at 1M (what the old
  snippet did, by conflating the declared context window with the
  compaction threshold) is what left sessions wedged near the limit.
• M3 stays documented as a one-flag opt-out
  (ccmm --model 'MiniMax-M3[1m]'), and the note now states what the
  function does not do: explicit --model arguments, custom and legacy
  model IDs pass through untouched, and your saved Claude Code settings
  are never edited.

Docs-only release — no code or packaged behavior changed.

Full Changelog:
voxtype-v0.3.7...v1.30.2