modelisted

# /mode — LLM Routing Switch Manage delegation across a multi-LLM setup. The pattern: Claude orchestrates, judges, reviews, and surgically fixes; cheaper or specialized CLIs handle the heavy lifting. | Layer | Typical CLI | Strength | | ----- | ----------- | -------- | | Heavy reading | Antigravity CLI (`agy`) | Large context window, fast cold reads. Replaces the Gemini CLI on consumer plans as of 2026-06-18; enterprise plans may still ship `gemini`. | | Heavy writing | Codex CLI | High-throughput code generation | | Surgical web search | Perplexity API (or others) | Live docs, breaking changes | Other skills in this collection read `.claude/llm-mode.json` to decide whether to delegate. When mode is `single`, every skill runs natively in Claude. ## Route by argument Parse the argument after `/mode`: - No args → **show status** - `multi` → **switch to multi** - `single` → **switch to single** - `stats` → **show delegation log** - `reset` → **reset to defaults** ## Show status (default) Read `.claude/llm-mode.json` and display: ``` ## LLM Mode: <MULTI / SINGLE> ### Delegation targets (in multi mode) READING (Antigravity CLI / agy): [available / not found] WRITING (Codex CLI): [available / not found] WEB SEARCH (Perplexity): [key set / key missing] ### Delegation stats Total: N successful delegations Failed: N (X% failure rate) Last: YYYY-MM-DD HH:MM via <skill> Switch with: /mode multi or /mode single Full log: /mo