mtor

Agent-first coding task dispatcher for Temporal workflows. Dispatches AI coding tasks to workers, reviews results automatically, and merges approved changes.

Install

uvx mtor
# or
pip install mtor

Quick start

# Dispatch a coding task
mtor "Add logging to server.py"

# Check status
mtor status ribosome-zhipu-abc123

# List recent runs
mtor list --since 24

# Health check
mtor doctor

Configuration

All config via environment variables:

Other settings (task queue, output paths) are configured in mtor/__init__.py.

Features

Auto-routing

Tasks are automatically routed to the best provider based on keywords:

• Exploration queries → droid
• Bug fixes → goose
• Build tasks → zhipu (default)

Override with --provider:

mtor "Fix the login bug" --provider goose

Risk classification

Tasks are classified as low / medium / high risk. High-risk tasks (deletions, config changes) are flagged in the dispatch response.

Spec file support

Pass a markdown file as the prompt — mtor reads it and strips YAML frontmatter:

mtor path/to/spec.md

Spec decomposition

Multi-task specs (with ## Task 1, ## Task 2 sections) are automatically split into individual workflows sharing the same preamble.

Dedup

Same prompt → same workflow ID. Temporal prevents duplicate dispatches natively.

Experiment mode

mtor -x "Try a new approach to caching"

Experiment tasks don't auto-merge — they stay on branches for manual review.

Deferred approvals

Gate high-risk tasks behind manual approval:

mtor approve ribosome-zhipu-abc123
mtor deny ribosome-zhipu-abc123

Commands

Every response is a JSON envelope with ok, command, result/error, and next_actions — designed for programmatic consumption by agents and tools.

Exit Codes

Requirements

• Python ≥ 3.11
• A running Temporal server
• A worker process executing coding tasks (e.g., ribosome + translocase)

License

MIT