Experimental
OpenAVO: open autonomy, fixed authority
OpenAVO is the project home for the open MetaHarness implementation of an Autonomous Variation Operator. The agent gets room to experiment; the harness keeps permissions, budgets, invariants, promotion, rollback and evidence.
GitHub activity: · Published:
What OpenAVO is
An Autonomous Variation Operator is the part of a self improving system that proposes the next change. OpenAVO lets an agent inspect, hypothesise, edit, execute, evaluate, repair, branch, restore and commit — while the harness decides what is allowed and what counts as evidence.
The framing in the repository is precise: the agent proposes the next experiment; the harness decides what may survive.
What changed
This commit establishes the canonical project home, separating the operator from the harness that governs it and pointing at the interactive explainer.
Get started
Node.js. The documented packages:
npm install @metaharness/avo agenticowExpected result: the variation operator available to a MetaHarness project. Read the explainer first — this is a control surface, not a utility.
No documented standalone MCP server. It operates inside a MetaHarness harness.
Use it today
Practical case: bounded automated repair. Input is a failing test; the workflow is hypothesise, edit, execute, evaluate under harness limits; the output is either a promoted fix or a clean rollback.
Acceptance test: give it a change that violates a protected invariant. The harness must block promotion regardless of how good the metric looks.
Push it further
Experimental commentary. Separating autonomy from authority is the only structure that makes self modifying systems reviewable. The limitation is invariant coverage: the harness can only protect what you thought to declare.
Falsifiable test: list your protected invariants, then ask what damage is still possible within them. That gap is your actual risk.