I’ve been experimenting with treating growth execution as a stateful system instead of a collection of tools.
GrowthClaw connects to OpenClaw as a skill and defines YAML workflows for context modeling, strategy generation, task planning, approval, execution, and evaluation. Tasks move through explicit states and every transition is policy-checked. All state is persisted in SQLite, so strategies and outcomes are replayable.
There’s also a scheduled strategy evolution loop that re-evaluates context and performance before proposing new work.
It’s essentially an attempt to treat growth as a feedback system with constraints rather than a series of prompts.
Code and architecture are here:
https://github.com/mrrkrieg/growthclaw
Top comments (0)