Claude Code

Claude Code is Anthropic’s official command-line agent. Agentslan dispatches it in headless mode (claude -p) and pipes prompts via the Claude Code CLI, so agents work without supervision once you wire them up.

The adapter type inside Agentslan is claude_local and the menu label is Claude Code (local).

Install Claude Code on the host

You need Node.js 18 or newer. Install Claude Code globally so the claude binary is on your PATH:

npm install -g @anthropic-ai/claude-code

Verify the install:

claude --version

If claude is not on your PATH, set the command field in the adapter configuration to the full path of the binary.

Sign in to Claude on the host

Run claude once interactively before connecting it to Agentslan:

claude

Complete the OAuth flow in your browser. After you sign in once, Claude Code caches credentials on disk and Agentslan can dispatch it headlessly.

You can also use an API key by setting ANTHROPIC_API_KEY in the host environment or in the adapter env map.

Add the agent in Agentslan

1. Open the Agentslan desktop app.
2. Go to Agents → Add agent.
3. Choose Claude Code (local) as the adapter type.
4. Set a model. The adapter ships with these options:
   • claude-opus-4-7 (Claude Opus 4.7)
   • claude-opus-4-6 (Claude Opus 4.6)
   • claude-sonnet-4-6 (Claude Sonnet 4.6)
   • claude-haiku-4-6 (Claude Haiku 4.6)
   • claude-sonnet-4-5-20250929 (Claude Sonnet 4.5)
   • claude-haiku-4-5-20251001 (Claude Haiku 4.5)
5. Save. The agent is now eligible to be assigned to issues.

Configuration reference

The fields below are read from the adapter’s agentConfigurationDoc constant in claude-local. Most defaults are sensible; surface them only when you need to override.

Field	Type	Notes
cwd	string, optional	Default absolute working directory for the agent process. Created if missing when possible.
instructionsFilePath	string, optional	Absolute path to a Markdown instructions file injected into the run at runtime.
model	string, optional	One of the model IDs above.
effort	string, optional	Reasoning effort, passed via --effort: low, medium, or high.
chrome	boolean, optional	Pass --chrome when running Claude.
promptTemplate	string, optional	Run prompt template.
maxTurnsPerRun	number, optional	Maximum agent turns in a single run.
dangerouslySkipPermissions	boolean, optional	Defaults to true. Passes --dangerously-skip-permissions so headless --print runs do not stall on interactive permission prompts.
command	string, optional	Defaults to claude. Override with a full path if Claude Code is installed somewhere else.
extraArgs	string[], optional	Additional CLI arguments.
env	object, optional	Extra KEY=VALUE environment variables for the run.
workspaceStrategy	object, optional	Currently supports { type: "git_worktree", baseRef?, branchTemplate?, worktreeParentDir? }.
timeoutSec	number, optional	Run timeout in seconds.
graceSec	number, optional	SIGTERM grace period in seconds before SIGKILL.

Notes

• Agentslan runs Claude in headless --print mode. Anything that would prompt for interactive input will hang the run; that is why dangerouslySkipPermissions defaults to true.
• Claude Code reads AGENTS.md from the working tree if present; that discovery cannot be suppressed.