spinosa-evolver

Applies scoped control-file and behavior-doc updates justified by evaluator audits.

7|Updated May 13, 2026
One-click install
npx skills add https://github.com/medialab/spinosa --skill spinosa-evolver-medialab
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: spinosa-evolver
Source: https://github.com/medialab/spinosa/tree/main/workspace-template/.opencode/skills/spinosa-evolver
Command: npx skills add https://github.com/medialab/spinosa --skill spinosa-evolver-medialab

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? After a research route completes, an evaluator audit may recommend changes to the framework's own control files and behavior docs. Without a governed process, such self-modifications risk touching source material, exceeding scope, or going unrecorded. This Skill applies only audit-justified edits within a strict mutation scope and records every change for future requests. ## Core Features & Use Cases - Audit-Gated Edits: Applies changes only when an evaluator audit report exists in agent_reports/ with an edit_recommended decision. - Scoped Mutation: Restricts edits to AGENTS.md, .agents/agents/, .agents/skills/, and behavior-defining docs under system/, refusing anything outside scope. - Evolution Reporting: Writes an evolution report to agent_reports/ documenting the triggering audit, files changed, rationale, and validation still required. - Use Case: An evaluator finds that an agent's instructions caused a recurring citation error. The evolver reads the audit, makes the smallest fix to the relevant behavior doc, and logs the change so the next request benefits. ## Quick Start Apply the framework edit recommended in the latest evaluator audit report and record the change in an evolution report.

Frequently Asked Questions about spinosa-evolver

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I update an AI agent's instructions based on evaluation feedback?▼

Run the evolver after an evaluator audit produces an edit_recommended decision. It reads the audit report, applies the smallest change to the targeted control files or behavior docs, and records the edit in an evolution report under agent_reports/.

What files can a self-evolving agent framework safely modify?▼

This Skill restricts mutations to AGENTS.md, the .agents/agents/ and .agents/skills/ directories, and behavior-defining docs under system/. It never edits raw source material, evidence packets, or completed answer reports.

Can the evolver change files without an evaluator audit?▼

No. An audit report with an edit_recommended decision is a hard prerequisite. If the audit is weak or the requested change exceeds the allowed scope, the Skill refuses to edit and explains why in the evolution report.

When do self-edits to agent behavior take effect?▼

Self-edits apply to the next request only, never retroactively to the current route. The current route's answer report and source corpus remain untouched.

What are the limitations of audit-driven framework evolution?▼

It cannot perform opportunistic cleanup, unrelated refactors, or out-of-scope changes, and it depends entirely on the quality of the evaluator's audit. Each edit still requires validation, which the evolution report flags as pending.