houyongsheng/deepseek-harness-molt
The coding agent that grows its own tools
MOLT is a coding agent that grows its own tools: after every job it reflects on what it did, and when it spots a reusable pattern it writes itself a new tool, tests it, and keeps it in a skill library so the next time the pattern appears the tool already exists. The README's walkthrough shows molt run "tidy the config files" --learn --mock growing parse_csv_line into .molt/skills/, then molt skills list and a later run reusing it. Skills only enter the library after passing their own test — the README calls this "honest self-modification" — and molt export / molt install let you share or pull someone else's toolbox (e.g. molt install git@github.com:you/toolbox.git).
Install
pip install molt-agentREADME quickstart, quoted verbatim: pip install molt-agent (the package is published on PyPI under that name — PyPI 0.2.0, verified 2026-09-12, homepage → github.com/houyongsheng/deepseek-harness-molt). The from-source alternative is documented too: git clone https://github.com/houyongsheng/deepseek-harness-molt, cd deepseek-harness-molt, pip install -e .. Then export DEEPSEEK_API_KEY=sk-... and run a task with molt run "add retry logic to the http client" --learn; with no API key the README documents --mock to run the same loop against a built-in fake model. Note this is a standalone CLI agent, not a dsh plugin bundle.
Compatibility
Python package installed via pip; runs on any OpenAI-compatible model (DeepSeek by default) and needs DEEPSEEK_API_KEY for real runs, or --mock for a keyless loop. The README says it is MIT-licensed with zero tracking. It is not a DeepSeek Harness plugin — it is a separate command-line coding agent in the same ecosystem.
Details
- Repo: houyongsheng/deepseek-harness-molt
- Category: Agent Capabilities
- Stars: 1
- Version: PyPI molt-agent 0.2.0 (verified 2026-09-12; project URL → github.com/houyongsheng/deepseek-harness-molt)
- Last push: 2026-08-14
- First seen: 2026-08-14
Recent updates
The README documents the compounding-toolbox model, the command set (molt run ... --learn, molt skills list, molt export / molt install, --mock), and a measure-and-publish section with accuracy reporting. It also covers trust and safety, a roadmap, and a Codex/Claude ecosystem section. The current PyPI release is 0.2.0 (verified 2026-09-12).
FAQ
- How do I install MOLT?
- pip install molt-agent from PyPI, or clone the repository and pip install -e .. Then export DEEPSEEK_API_KEY — or use --mock to run the same loop against a built-in fake model without a key.
- Is this a DeepSeek Harness plugin?
- No — the README describes a standalone command-line coding agent that runs on any OpenAI-compatible model. It installs with pip, not dsh plugin add.
- How do its self-written tools stay safe?
- Per the README, a skill only enters the library after passing its own test — the agent can grow itself new tools, but they must prove they work first.
Alternatives
tt-a1i/archify · foryourhealth111-pixel/Vibe-Skills · Anionex/agent-vision-toolkit