DEDUPED Hot AI repos, turned into use cases Get 5 a day

How to draft an accept-or-reject rule for harness updates with Harness RSI

Write a checklist that accepts a harness change only when it beats the current state on held-out evidence.

  • Research
  • Advanced
  • 60 minutes
  • From the README
  • Any agent

Who it is forAn engineer running a self-updating agent who needs a rule for when a proposed harness change is kept and when it is discarded.

The problemWithout a fixed rule, a self-updating harness can keep changes that look good on a few runs but do not help overall.

The outcomeA five-item checklist that ties acceptance to an external evaluator, a held-out batch, and a logged decision for every candidate.

What you need first

  • A web browser to open the README, the project page, and the linked PDFs
  • A goal to pick an update target, classify a paper, or cite the survey

Steps

  1. Read the Definition section and note that self-improvement means a beneficial change established through evaluation.
  2. Open the Evaluation section and list the evaluation practices it describes for comparing candidates.
  3. Check the Reported Evidence section to see how published results describe their comparisons.
  4. Write a checklist: external evaluator, held-out batch, a rule for accepting a candidate, a rule for rejecting it, and a log entry for each decision.
  5. Run one candidate change through the checklist and confirm that a rejected candidate leaves the harness unchanged.

Prompt for your agent

Using the zgcagi/Harness-RSI repo at https://github.com/zgcagi/Harness-RSI, read the Definition and Evaluation sections and write a five-item checklist that accepts a harness update only when it is established through evaluation for <my agent project>. Output the checklist as numbered lines, each with the README section it comes from. Verify that every item names an evaluator outside the candidate's editable surface.

Caveat. The repo states the acceptance principle but does not supply a ready-made evaluation harness, so you must build the evaluator yourself.

Share: X LinkedIn Hacker News Email Copy link

Disclaimer

Deduped lists open-source repositories that belong to other people. We do not own, maintain, audit, or vouch for any of them, and a repo appearing here is not an endorsement or a safety check. The use cases, steps, and prompts are written by AI and may be incomplete, outdated, or just wrong. Where a use case is marked "Seen in the wild", we link to public posts where someone described something similar, but we have not verified those posts and the use case itself is still our AI's own write-up. Use all of this at your own risk. Only run code, commands, or agents on systems and accounts you are authorized to use, and read the code and every command before you run it, especially anything that touches credentials, files, or the network. Nothing here is professional, security, legal, or financial advice. All trademarks and project names belong to their respective owners, and their use here does not imply affiliation. Content is provided as is, without warranties of any kind. Found a mistake, a safety problem, or something you want removed? Email [email protected] and we will look at it promptly. Full disclaimer page