MOLT v0.2.0 — roadmap done, Codex & Claude integration

MOLT v0.2.0

Roadmap complete, plus Codex / Claude ecosystem integration.

New in 0.2.0

• molt skills dedup — remove shadowed copies, flag identical code
• molt eval <name> — accuracy benchmark against cases.json
• molt publish --remote — push the toolbox to a git registry
• molt run --stream / --sandbox — live output, subprocess skill isolation
• molt export --format claude|codex — your grown tools work in Claude Code / Codex
• new built-ins list_dir, search; 16 passing tests

Verified

A MOLT-grown skill (parse_kv_records) exported to .claude/skills/ was
actually used by Claude Code to solve a task correctly.

MOLT v0.1.0 — the coding agent that grows its own tools

MOLT v0.1.0

First public release. After each task, MOLT reflects, authors a reusable tool, tests it, and keeps it in a skill library that compounds.

Highlights

• molt run "task" --learn — run a task and grow a tool
• molt evolve "goal" -n N — loop, watch the toolbox accumulate
• molt skills list/show/remove, molt export/install — manage & share skills
• --mock — the whole self-evolving loop runs with no API key
• 9 passing tests, keyless CI-ready

Verified demo

A real DeepSeek-model run grew parse_kv_records (tested → committed → reusable).

Install

git clone https://github.com/houyongsheng/deepseek-harness-molt
cd deepseek-harness-molt
pip install -e .
export DEEPSEEK_API_KEY=sk-...
molt run "tidy the config files" --learn