Self-improving template harness

The idea

After work finishes, two proposers edit the template's harness from the run ledger. Each candidate is probed on the same request. A composer and a blind selector pick one, which a person adopts (or auto-adopt does, if the checks pass).

The problem
A harness block stays as written even when finished work shows what goes wrong.
Why it matters
Edits come from evidence, stay within a budget, and roll back after three failures.
Where it stands
Partly built as of

A step-by-step replay of this idea is not published yet. The library adds one when a recorded run shows it.

  • Opinionated harness per work type

    Each work template carries steps, a quality bar, pitfalls, tool guidance and practitioner notes, plus a minimum model tier and effort. Its named industry checklists are settled by detectors or by attestations with evidence.

  • Keep-or-discard harness loop

    A harness change is kept only if the declared capability metrics stay within tolerance, one efficiency metric improves by the minimum, and frozen held-out cases confirm it. Campaigns count every attempt.

  • Worker insights and verified work per token

    Reports built from the call records show how each worker spends time, tokens, context, models and tools. The Performance tab shows verified results beside the tokens they took.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.