acp

Spawn external coding agents via the Agent Client Protocol over stdio.

1|Updated Mar 30, 2026
One-click install
npx skills add https://github.com/Logarn/Worklin-ai --skill acp-logarn
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: acp
Source: https://github.com/Logarn/Worklin-ai/tree/main/assistant/src/config/bundled-skills/acp
Command: npx skills add https://github.com/Logarn/Worklin-ai --skill acp-logarn

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires @agentclientprotocol/claude-agent-acp, @zed-industries/codex-acp, @google/gemini-cli.

What problem does it solve? Delegating substantial coding tasks to external agents like Claude Code, Codex, or Gemini normally requires manual terminal setup and monitoring. This Skill lets the assistant spawn, steer, and track those agents as subprocesses speaking the Agent Client Protocol, streaming results back into the conversation. ## Core Features & Use Cases - Agent Spawning: Launch Claude Code, Codex, or Gemini as ACP subprocesses with a task and working directory, with automatic one-time adapter installation when binaries are missing. - Session Management: Check status, abort running sessions, and steer in-flight or completed sessions with new instructions, resuming from persisted history when supported. - Secure Credential Handling: Store Claude OAuth tokens and Gemini API keys in the credential store so secrets never appear in chat or workspace config. - Use Case: Ask the assistant to hand a large refactor to Claude Code in a git worktree, then check on it later and steer it with follow-up instructions once it completes. ## Quick Start Ask the assistant to spawn a Claude Code agent to implement a specific feature in the current project directory.

Frequently Asked Questions about acp

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I spawn a Claude Code agent from the assistant?▼

Use the acp_spawn tool with a task description and optional working directory. The claude-agent-acp adapter is installed automatically on first use, and the agent runs as a subprocess streaming results back into the conversation.

What coding agents can be spawned via ACP?▼

Three agents ship out-of-box: claude via the claude-agent-acp adapter, codex via the codex-acp adapter, and gemini via gemini --acp, which speaks ACP natively. Natural names like 'claude code' or 'gemini cli' resolve to these canonical ids.

How do I resume or follow up on a completed ACP session?▼

Use acp_steer with the session id and a new instruction. Completed sessions are transparently resumed from persisted history via ACP session loading when the agent supports it; otherwise the error explains why and you can fall back to acp_spawn.

Does the Gemini CLI need a separate ACP adapter?▼

No, the Gemini CLI speaks ACP natively when launched with the --acp flag, so no separate adapter binary exists. The CLI itself is installed from @google/gemini-cli when missing, and a Gemini API key can be stored in the credential store.

Why does spawning a Codex agent fail even after the adapter installs?▼

The codex-acp adapter shells out to the underlying codex CLI, which must be on PATH separately at version 0.111 or higher. You also need to authenticate via codex login, CODEX_API_KEY, or OPENAI_API_KEY.

When should I not use an external ACP coding agent?▼

Avoid spawning an external agent when the task is small enough to handle inline with the assistant's own tools. ACP is meant for substantial autonomous coding work where the agent needs its own file editing and terminal access.