← ClaudeAtlas

crosstalklisted

Consult Antigravity (Gemini) from Claude Code without blocking, and route the answer to whichever of your sessions needs it. Use for wide-context reads a second model should carry (large codebase sweeps, long log or transcript analysis, broad web research), for a genuine second opinion on a design or a diagnosis, and for long-running investigations you do not want to sit and wait for. Also use when the user says "ask Gemini", "ask Antigravity", "ask agy", "second opinion", "get another model on this", or "offload this". Not for work that needs this session's uncommitted context — the other model cannot see it.
obeskay/crosstalk · ★ 1 · AI & Automation · score 72
Install: claude install-skill obeskay/crosstalk
# crosstalk Antigravity (Gemini) runs as a separate CLI (`agy`). This skill makes it a peer you consult, with the thread kept across calls, and the answer delivered where it is useful. `crosstalk.sh` sits next to this file. Call it with an absolute path — resolve it from this skill's own directory rather than assuming a working directory. ## Decide first: is this worth another model? Consult Gemini when at least one is true: - **Context you would rather not spend.** A 200-file sweep, a 50k-line log, a long transcript. The answer matters; the raw material does not need to live in this conversation. - **A second opinion has real value.** Architecture calls, ambiguous bug diagnoses, "did I miss something" reviews. Two models disagreeing is a signal worth having. - **It runs long and you have other work.** Delegate it, keep going, read the answer when it lands. Do not consult when the task depends on what only this session knows — uncommitted edits, the user's last three corrections, the plan you just agreed on. Gemini starts cold. Sending it a question it cannot answer wastes a call and produces confident nonsense. ## Ask ```bash /abs/path/to/skills/crosstalk/crosstalk.sh ask "<prompt>" --thread <name> ``` `--thread <name>` resumes the same Gemini conversation next time you use that name, so a follow-up does not re-explain the problem. Use one thread per topic (`auth-refactor`, not `question-3`). Drop the flag for a one-off. Other flags: `--model`, `--effort low|medium