agenthatch
Compile markdown skills into standalone Python agents
TLDR
SYNOPSIS
agenthatch command [arguments] [options]
DESCRIPTION
agenthatch is a compiler that turns a markdown SKILL.md definition into a standalone, runnable Python AI agent. It parses the skill file, runs an AI inference pass to infer tools and structure, and generates a complete Python package with its own CLI entry point, type-annotated tool definitions, and MCP integration.Registered skills live in a local index (the "skillhouse"). The typical flow is to register a skill, hatch it to generate code, then run the resulting agent in an interactive terminal UI.
PARAMETERS
--no-generate
Skip the code-generation phase so the inferred spec can be reviewed first (used with hatch).--force
Overwrite an existing hatched agent.--dry-run
Preview the output without writing files.
COMMANDS
init
Initialize configuration and set up the LLM provider.skills add path
Register a SKILL.md file in the skillhouse index.skills list
Display all registered skills.skills delete name
Remove a skill from the skillhouse.hatch name
Run the full compilation pipeline (parse, AI inference, code generation).run name
Launch a hatched agent in interactive terminal UI mode.search query
Query the skillhouse index.assemble
Rebuild an existing skillhouse agent.doctor
Diagnose the environment and dependencies.
CAVEATS
Requires Python 3.11 or later. Compilation calls an LLM provider (OpenAI, DeepSeek, Anthropic, or any OpenAI-compatible endpoint), configured via a TOML file, so an API key and network access are needed for the hatch step.
HISTORY
agenthatch is written in Python and released under the MIT license. It was published in 2026 around the idea of treating markdown skill files as source code that compiles into deployable agents.
