acp

ACP agent orchestration - spawn external coding agents (Claude Code, Codex) to work on tasks via the Agent Client Protocol. Each agent runs as its own subprocess speaking ACP over stdio and streams results back into the conversation.

## Usage

Use `acp_spawn` to delegate a coding task to an external agent. The agent runs as a subprocess speaking the ACP protocol over stdio and streams results back.

Users can refer to agents by natural names: "claude code", "codex cli", and "openai codex" all resolve to the canonical `claude` and `codex` ids (unless the user's config defines an agent literally keyed by that name, which always wins).

## First-time setup

ACP is always available - default profiles for `claude` and `codex` ship out-of-box, so no config edit is needed to start. First-time setup is just making the adapter binary available, then spawning:

1. Install the adapter binary if it's missing. This happens automatically: when `acp_spawn` finds the agent's binary missing from PATH, the assistant installs it once via a sandboxed bun global install and proceeds in the same call (see "Automatic adapter availability" below).

2. Call `acp_spawn`. Do NOT run `vellum sleep && vellum wake` - that kills the conversation.

## Automatic adapter availability

When `acp_spawn` finds the agent's binary missing from PATH, the assistant installs it once via a sandboxed bun global install and then runs the real installed binary. The install runs in a fresh empty temporary directory (never the task's project directory), with known secrets stripped from the installer environment and the registry pinned to the public npm registry, so a malicious project directory cannot hijack package resolution or capture a token. After this one-time install, the adapter is a normal trusted binary on PATH and every later spawn (and resume) uses it directly.

Only the allowlisted out-of-box packages are ever installed this way (`@agentclientprotocol/claude-agent-acp`, `@zed-industries/codex-acp`); user-configured agents with custom commands are never installed automatically.

Manual installation is fallback guidance for unusual setups: bun unavailable, restricted global installs, or an auto-install failure (the failure reason is surfaced in the tool result).

```bash
bun add -g @agentclientprotocol/claude-agent-acp   # claude
bun add -g @zed-industries/codex-acp               # codex
```

## Claude setup

The `claude-agent-acp` adapter requires a Claude **OAuth token** (`sk-ant-oat…`), NOT an API key (`sk-ant-api…`). Every spawn injects the stored token as `CLAUDE_CODE_OAUTH_TOKEN` automatically. The write path rejects an API key in this field, so never direct a user to paste an `sk-ant-api…` key here.

**Primary: the in-app Connect Claude Code flow.** When a spawn fails because the token is missing, the UI **automatically renders an inline "Connect Claude Code" card** for the failed step — one click on desktop (loopback), one paste on cloud. The card restores itself after a reload or reconnect, so it stays available. It mints and stores the OAuth token so it never enters the conversation or the workspace config.

**When a spawn fails for a missing token, do NOT prompt or instruct the user yourself.** The inline card already handles it, and the task **auto-continues** once they connect — so you don't need them to re-ask. Specifically, do NOT tell them to run `claude setup-token`, run `assistant credentials set`/`prompt`, open a terminal, or paste an `sk-ant-oat…` token — and do NOT retry the spawn yourself. Add at most one short sentence pointing at the card ("Click **Connect Claude Code** to sign in — I'll pick it back up once you're connected"), then stop and wait. Keep it terse and never say where the card is (no "above"/"below"/"at the bottom" — its placement is a UI detail you can't see): do not narrate that a card appeared, explain how the sign-in works, or say "nothing to paste" (the cloud flow does paste a key).

**Fallback (headless environments where no inline card can appear):** the user runs `claude setup-token` on a machine where they are logged in to Claude, then stores the result via the secure prompt:

```bash
assistant credentials prompt --service acp --field claude_oauth_token --label "Claude OAuth Token"
```

This is strictly for headless/channel sessions. In an interactive session the daemon **refuses** this prompt for `acp/claude_oauth_token` and returns a message pointing at the inline Connect card, so do not run it to work around the card — just wait for the user to connect.

Do NOT ask the user to paste the token into chat — the secure prompt keeps it out of the conversation and the workspace config.

## Codex setup

The `codex-acp` adapter is installed automatically when missing, but it shells out to the underlying `codex` CLI, which must also be on PATH:

1. **Install the Codex CLI** (version 0.111 or higher) via OpenAI's distribution channel of choice. The adapter will fail if `codex` isn't on PATH.

2. **Authenticate.** The `codex-acp` adapter inherits whatever auth the underlying `codex` CLI uses. Typical flows:
   - `codex login` (OAuth)
   - `CODEX_API_KEY` environment variable
   - `OPENAI_API_KEY` environment variable

Do NOT put API keys (or any secret) in the workspace config file - secrets never belong in the workspace directory. Use the credential store instead.

## Critical: correct agent command

- Two agents are supported out-of-box: `claude` (via the `claude-agent-acp` adapter) and `codex` (via the `codex-acp` adapter).
- NEVER use `claude`, `claude -p`, `claude --acp`, or the bare `codex` CLI as the ACP `command`. Claude and Codex only speak the protocol through their dedicated `*-acp` adapters.
- Default profiles for both ship out-of-box. Users only need an `agents.<id>` entry in config if they want to override the defaults (e.g. point to a custom binary path or pass extra args/env). An `acp.agents.<id>` entry replaces the bundled default entirely (no field merge), so any ove