v0.3.0 — rules + 32 adapters + subagent-follow switching

New

• ccse rules — inject/remove a behavioral snippet (built-in caveman + rtk) into every agent's global instructions file, idempotent via marked block. Same fleet-wide lever --model gives for models, but for behavior.
• 8 new adapters (25 → 32): openakita jcode dsh openclaude openhands commandcode + model-slot extensions. See README table.
• --model now covers subagents: opencode agent.{build,explore,general,plan}, openclaw subagent, snow basicModel, claude CLAUDE_CODE_SUBAGENT_MODEL — each keeping its route prefix. Claude tiers stay independently settable.
• codex: literal api_key providers supported alongside env_key; switching to a non-OpenAI gateway auto-releases requires_openai_auth (which would otherwise brick codex with OAuth).

Fixed

• Windows: an agent config that exists but can't be parsed (UTF-8 BOM written by Windows tools, corrupt TOML) no longer aborts the whole switch — the adapter is skipped with a one-line warning (ccse: skip <id>: unreadable config (...)), and BOM'd TOML now parses via utf-8-sig.
• jcode: stale default_provider falls back to the first providers block; model writes mirror into the provider block + [[...models]] registry.

Install

git clone https://github.com/Tonystarkw12/cc-switch-enhanced
cd cc-switch-enhanced && uv tool install . --force   # or pipx install .

Windows users on 0.2.4 hit by the EmptyKeyError crash: upgrade and re-run — the offending agent will be skipped with a warning instead of a traceback.

ccse 0.1.2

claude --model now also sets CLAUDE_CODE_SUBAGENT_MODEL; envrc no duplicate-append when already in sync.

ccse 0.1.1

snow adapter added; claude model names auto-carry [1M] suffix (glm-5.2 -> glm-5.2[1M]); empty model slots shown as .

ccse 0.1.0

First release. One-line ccse --model NAME across 16 coding agents. Auto-snapshot + ccse undo. See README.