Agents & Subagents

Hanzo Dev can launch external CLI agents and orchestrate them in multi-agent subagent flows such as /plan, /solve, and /code.

Agent Configuration

Configure agents in config.toml:

[[agents]]
name = "code-gpt-5.2-codex"
command = "coder"
args = ["--foo", "bar"]
args_read_only = ["-s", "read-only", "-a", "never", "exec", "--skip-git-repo-check"]
args_write = ["-s", "workspace-write", "--dangerously-bypass-approvals-and-sandbox", "exec", "--skip-git-repo-check"]
env = { CODE_FOO = "1" }
read_only = false
enabled = true
description = "Frontline coding agent"
instructions = "Preamble added to this agent's prompt"

Fields

Field	Description
`name`	Slug/alias shown in pickers
`command`	Executable path (defaults to name)
`args`	Base arguments
`args_read_only`	Override args for read-only mode
`args_write`	Override args for write mode
`env`	Environment variables
`read_only`	Force read-only even if session allows writes
`enabled`	Show/hide in pickers
`description`	Agent description
`instructions`	Preamble for agent prompt

When no [[agents]] are configured, Hanzo Dev provides: code-gpt-5.2, code-gpt-5.2-codex, claude-opus-4.5, gemini-3-pro, code-gpt-5.1-codex-mini, claude-sonnet-4.5, gemini-3-flash, claude-haiku-4.5, qwen-3-coder.

Subagents

Configure multi-agent slash commands:

[[subagents.commands]]
name = "plan"
read_only = true
agents = ["code-gpt-5.2-codex", "claude-opus-4.5"]
orchestrator_instructions = "Guidance for orchestrator before spawning agents"
agent_instructions = "Preamble added to each spawned agent"

The orchestrator fans out agents, waits for results, and merges reasoning.

TUI Controls

/agents opens the settings overlay to the Agents section
Agent editor: create or edit agents (enable/disable, read-only, instructions)
Subagent editor: configure per-command agent lists for /plan, /solve, /code

AGENTS.md and Project Memory

Hanzo Dev loads AGENTS.md files along the path (global, repo root, cwd) up to 32 KiB total. Contents become system/developer instructions on the first turn.

Headless Execution

# Stream JSONL events
dev exec --json "your prompt"

# Structured JSON output
dev exec --output-schema schema.json --output-last-message out.json "your prompt"

# Read-only by default; add --full-auto for edits
dev exec --full-auto "your prompt"