skills/gemini-use-claude-acp/SKILL.md
Delegate a sub-task to Claude Code via the Agent Client Protocol (ACP). Use this skill whenever you want to hand off work to Claude — complex agentic coding with MCP tool access, detailed multi-file refactors, tasks requiring Claude's reasoning style, or anything where Claude's strengths give an advantage. Also invoke when the user asks you to "ask Claude", "use Claude for this", or "run this through Claude". The script handles subprocess lifecycle and ACP session setup; you just provide the prompt and read stdout.
npx skillsauth add baphomet480/claude-skills gemini-use-claude-acpInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Spawn a Claude Code ACP adapter subprocess, send a prompt, stream the response to stdout, and terminate cleanly. No persistent session — one prompt, one response.
cd ~/.gemini/skills/gemini-use-claude-acp/scripts
pnpm install
This installs @agentclientprotocol/sdk into the local node_modules/. Subsequent runs skip this step.
CLAUDE_MODEL=claude-sonnet-4-6 \
TARGET_CWD=/path/to/project \
node ~/.gemini/skills/gemini-use-claude-acp/scripts/claude-delegate.mjs \
"Refactor the auth module to use JWT tokens. The code is in src/auth/."
The response streams to stdout as Claude generates it. Capture it:
response=$(node ~/.gemini/skills/gemini-use-claude-acp/scripts/claude-delegate.mjs "your prompt")
| Variable | Default | Description |
|---|---|---|
| CLAUDE_MODEL | claude-sonnet-4-6 | Model to use (applied via ACP unstable_setSessionModel) |
| TARGET_CWD | current working dir | Working directory passed to the Claude session |
pnpm dlx @zed-industries/claude-code-acp with:
CLAUDE_CODE_SKIP_PERMISSIONS=1 — auto-approves all tool callsTERM=dumb — prevents ANSI escape sequences in outputCLAUDECODE unset — prevents the adapter refusing to start inside another Claude session~/.local/bin prepended to PATH — ensures the claude binary is found@agentclientprotocol/sdk's ClientSideConnection + ndJsonStreaminitialize → newSession → unstable_setSessionModel → prompt over ACPagent_message_chunk deltas to stdout as they arriveIf the script exits non-zero, check stderr for:
pnpm dlx download failures (network, auth)claude CLI is authenticatedCLAUDECODE environment variable being set (indicates you're running inside Claude Code; the adapter refuses to nest)The script does not retry — surface the error to the user and suggest re-running after fixing the root cause.
development
Sets up, configures, and optimizes Google Analytics 4 (GA4) properties. Evaluates websites for proper GA4 implementation, tracking codes, and configuration improvements. Uses the Google Analytics Admin API for programmatic setup or provides manual integration paths via gtag.js or Next.js Third Parties.
development
Open-source intelligence on people, companies, domains, and B2B accounts. Use when the user wants to investigate, vet, research, or build a dossier on a target — phrases like "OSINT", "due diligence", "background check", "research this person", "look into [company/domain]", "vet this prospect/vendor", "what does X do", "is this account worth pursuing", "find me a contact at", "who's the buyer for", or any open-source investigation task. Disambiguates identities before reporting and grades every claim by independent source count.
development
Generate, edit, describe, restyle, restore, thumbnail, and batch-process images using xAI (Grok) or OpenAI image APIs and GPT-4o vision. Default provider is xAI ($0.02/image flat rate). Use this skill whenever the user asks to generate, create, make, draw, or design an image or picture using AI, or wants to edit, modify, transform, restyle, composite, or inpaint an existing image. Also handles image description and alt-text generation, background removal, style transfer, photo restoration, thumbnail creation, and batch generation from JSON manifests. Trigger when the user mentions DALL-E, gpt-image, Grok image, xAI image, OpenAI image generation, or wants AI-generated visuals for any purpose (logos, mockups, illustrations, thumbnails, icons, concept art, memes). Also trigger for batch image generation, generating a set or series of images, processing multiple images from a manifest, or creating consistent image collections. If the user says "make me an image of...", "generate a picture", "edit this photo to...", "describe this image", "remove the background", "make this look like watercolor", "restore this old photo", "create a thumbnail", "generate a batch of images", or "process this image manifest", this is the skill to use.
testing
Agentic OS Orchestrator. Process and execute tasks from the shared .agent/state/tasks.json queue. Use when the user asks to 'check the queue', 'process tasks', or run the heartbeat.