Skip to content
Development
Skill

/operate-acp-agent-integration

Install, launch, validate, and diagnose existing ACP agents across Zed, Xcode, and other clients. Use for registry discovery, custom launch commands, auth, capabilities, sessions, and logs.

From plugin
socket
9200 skills5 MCP

Claude Code

$ npx -y skills add gaelic-ghost/socket --skill operate-acp-agent-integration --agent claude-code

SKILL.md

operate-acp-agent-integration.SKILL.md
name: operate-acp-agent-integration
description: Install, launch, validate, and diagnose existing ACP agents across Zed, Xcode, and other clients. Use for registry discovery, custom launch commands, auth, capabilities, sessions, and logs.
metadata:
  hermes:
    category: agent-portability
    tags: [acp, zed, xcode, hermes, diagnostics]

Operate ACP Agent Integration

Operate an existing ACP agent without confusing client setup, agent-native configuration, and protocol behavior. Read `references/acp-operations-map.md` for the current connection and failure map.

Preflight

1. Name the ACP client and ACP agent/server process. 2. Treat ACP v1 as the current latest protocol and v2 as draft design work. Check the capabilities negotiated by the actual pair; do not infer wire compatibility from SDK package versions or implement a draft RFD as stable. 3. Check the canonical ACP Registry with the managed ACP registry checker. 4. If the agent is missing, use its official local executable only when the client supports custom agents. Keep registry absence distinct from missing ACP support. 5. Run the agent's documented non-interactive health or version check before opening the client.

For Hermes, check `hermes acp --version`, `hermes acp --check`, `hermes doctor`, and `hermes status` as needed. ACP reuses Hermes provider credentials, config, skills, and state; it does not create a separate provider setup.

Connect And Validate

1. Configure the client launch command and environment with the smallest required surface. 2. Confirm stdout is reserved for JSON-RPC and human-readable logs go to stderr. 3. Inspect the initialization response for protocol version, agent capabilities, authentication methods, and implementation identity. 4. Complete agent-owned or terminal authentication before starting a session. 5. Create a harmless session and verify working-directory binding. 6. Verify streaming messages, tool calls, file diffs, terminal rendering, permission requests, cancellation, and any advertised load, resume, fork, list, close, mode, or config-option behavior independently. 7. Verify forwarded MCP servers separately from the agent's native MCP config.

Completed v1 additions include generic session configuration options, `session/list`, session-information updates, `session/resume`, and `session/close`. Elicitation, boolean config-option refinements, `session/delete`, logout, additional directories, remote transports, and the v2 prompt lifecycle remain draft RFDs unless the live ACP documentation says they have advanced. Gate experimental behavior on explicit capabilities and label it as draft in reports.

Host Handoffs

  • For Zed registry/custom-agent settings and ACP logs, use `operate-zed-agent`.
  • For Xcode's Add an Agent flow, Xcode-owned tools, and permissions, use

`apple-dev-skills:xcode-coding-intelligence-workflow`.

  • For an external agent running outside Xcode and calling Xcode tools, stop:

that is MCP through `xcrun mcpbridge`, not an ACP-hosted Xcode agent.

  • For a custom program driving Hermes, reconsider the TUI gateway or API server

through `choose-agent-integration-protocol`.

Diagnose By Layer

  • Discovery failure: inspect the canonical registry or custom client config.
  • Spawn failure: inspect executable path, arguments, environment, and stderr.
  • Handshake failure: inspect protocol version and capabilities.
  • Authentication failure: inspect advertised auth methods and agent-native

provider state.

  • Empty or wrong workspace: inspect session cwd and additional roots.
  • Missing tool: inspect client-forwarded MCP and agent-native MCP separately.
  • Missing history: inspect whether load, resume, list, or fork is advertised;

do not assume every client exposes every optional method.

Report

Report registry status, launch source, client and agent versions, negotiated protocol and capabilities, auth owner, cwd, session behavior, MCP paths, permission behavior, evidence collected, and unresolved client-specific UI checks.

Read more
@gaelic-ghost
Ships withsocket

Stuff for Agents on macOS Promo audio: Socket Codex Marketplace Promo

9200 skills5 MCP
Get the whole plugin

Stars on GitHub

Momentum

stars in total
9
Star history of gaelic-ghost/socket on GitHub, a chart by star-history.com

Other skills on socket.

orchestrate-agent-work

orchestrate-agent-work

Coordinate bounded worker tasks with a launch envelope, report-back, escalation, and synthesis contract. Use before spawning,…

@gaelic-ghost@gaelic-ghostInstall Skill

More skills from other plugins.

brainstorming
Auto-invoked

brainstorming

You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or…

@obra@obraInstall Skill
ask-matt

ask-matt

Ask which skill or flow fits your situation. A router over the skills in this repo.

skills10d309k
@mattpocock@mattpocockInstall Skill
accessibility
Auto-invoked

accessibility

Design, implement, and audit accessible UI to WCAG 2.2 Level AA across Web, iOS, and Android — semantic ARIA roles and labels,…

ecc12d293k
@affaan-m@affaan-mInstall Skill
compiler-commit

compiler-commit

Use when you want to verify compiler changes and commit with the correct convention. Runs tests, lint, and format, then commits…

react1d252k
@facebook@facebookInstall Skill

Questions about operate-acp-agent-integration.

Which coding agents can use operate-acp-agent-integration?
operate-acp-agent-integration installs in Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and Windsurf with the skills CLI, using each agent's flag shown on this page. Flowy's auto-invocation, which fires a skill when your prompt matches, runs in Claude Code only.