Self-Improving Ecosystems skill
An open kit of contracts, schemas and a skill for systems where an evaluator, not a schedule, decides what improved.
- 64 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Scripts create scaffolds and download external AVO-lite source code without executing it.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/self-improving-ecosystemsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Scripts create scaffolds and download external AVO-lite source code without executing it.
This is third-party code. Review the repository files before installing.
What it does
The repository describes how to build self-improvement loops with verifiable steps. The principles: correctness and quality are scored separately, candidates are isolated from canonical state, a likely winner passes adversarial verification, every attempt keeps evidence and promotion follows Git and deployment facts. It ships the `self-improving-ecosystem` skill, schemas, examples and small scripts on the Python standard library: an installer, scaffold generator, manifest and policy validator and progress check. Three loop types: optimisation against a scalar objective, maintenance of a work queue and observation that produces proposals. It is a reference kit, not a ready autonomous agent.
Who it is for. Engineers designing agent optimisation and maintenance loops who want the evaluator, rollback and authority boundaries thought through up front.
Good fit when
- A loop is being designed where the agent proposes changes and an evaluator accepts or rejects them.
- An existing automation is audited for weak spots such as no rollback or no external stall detection.
- A team scaffolds an ecosystem with the `local`, `project` or `live` profile.
- A manifest and policy should be validated by a script before launch.
Not a fit when
- A ready autonomous agent that works right after install is needed.
- The task is one-off and needs no evaluation loop.
- There is no way to define an evaluator and acceptance criteria.
Example request
Audit the pull request queue maintenance loop with the self-improving-ecosystem skill and list gaps in the evaluator and rollback.Limitations
This is a reference kit, not a ready system: generated commands fail closed until real sensors and adapters are configured. The AVO-lite installer fetches the latest default branch by default and prints the commit, so inspect the checkout and pin a version with `--ref` before running. It needs Git, Python 3.10 and a POSIX shell. Deployment, external deadman switches and alerts are yours to implement.
How to disable. Delete the cloned repository folder and any scaffolds created from the templates. Remove the skill from the agent skills directory.
FAQ
Is this a ready autonomous agent?
No. The README calls it a reference kit: contracts, schemas, skills and examples, while you write the sensors and adapters.
Which loop types exist?
Three: optimisation against a scalar objective behind a hard correctness gate, queue maintenance and observation that yields proposals instead of code.
How do I create a scaffold?
Run `./scripts/init-ecosystem.sh ../my-ecosystem my-ecosystem project`. The `local`, `project` or `live` profile sets requirement strictness.
What does `validate.py` check?
The manifest, policy and example event stream of the created ecosystem.
Where does AVO-lite come from?
`scripts/install-avo-lite.sh` fetches the source without executing it and prints the exact commit. Pin a version with `--ref`.
Discussion
No questions or reviews yet.
Related
SkillsSelf-Improving Agent
self-improving-agent
Skill for OpenClaw: the agent logs errors, corrections and missing features to .learnings files and promotes verified lessons into workspace memory
A CLI that runs your agent on tasks and revises its skills, keeping a change only when held-out tasks confirm the gain
Parallel self-evolution of skills and prompts: workers propose edits and an aggregator merges them into one versioned artifact
A skill that designs, tests, evaluates and packages other agent skills, starting from architecture
Agent self-improvement framework: optimizes prompts, memory, skills or harness code with method and evaluation set separately.
A skill that creates and reviews other skills, weighing usefulness, trigger boundaries, setup and stability before release.
README in short
A reference kit for designing self-improving systems: evaluator, candidate isolation, evidence and rollback. The skill helps audit and extend such loops.