Skip to content
Get packs

Harness Evolve

Refines an existing agent harness using feedback from past runs and logs changes

Source Download skill
Repository
revfactory/harness
Stars
9,305
Licence
Apache-2.0
Files
1 file

What it does

This skill teaches an AI agent to improve a multi-agent setup (called a harness) that it has already built for a project. It gathers feedback on past results, finds what worked and what did not, and turns that into general rule changes to agents, skills, or the orchestrator. Each change is recorded in a change-history table so future setups start from a better baseline.

Great for

  • Developers who built an agent harness with Claude and want to tune it after real use
  • Teams that want a documented history of why their agent setup changed
  • Users who notice repeated fix requests or failed retries in their agent workflow

Skip it if

  • Building a new harness or redesigning its architecture; the SKILL.md says the separate harness skill handles that
  • Anyone wanting one-off content edits, since the skill asks for general principles rather than quick patches

Try asking

  • “The report came out too shallow, so update the harness so the analysis agent goes deeper”
  • “Save this feedback into the harness: the orchestrator should validate before drafting”
  • “Summarize the harness lessons learned and update the change history”
  • “The harness trigger missed my request, so expand the skill description”
Step-by-step guide: how to use Harness EvolveInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the Apache-2.0 licence and an attribution note.

    evolve.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip evolve.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip evolve.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Give concrete feedback about results so the skill can generalize it into a principle
  • Keep changes to one at a time, as the skill recommends, so you can see which change helped
  • Check the change-history table in CLAUDE.md after each evolution round

What's inside

  • SKILL.md

Redistributed unmodified from revfactory/harness under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.