improve-loop SKILL.md skillA
improve-loop from ai-evals-course/cartwheel-homeworks is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add ai-evals-course/cartwheel-homeworks/improve-loop@git:20260930.c1b179fPins git:20260930.c1b179f in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 17 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- ai-evals-course/cartwheel-homeworks
- Path
.agents/skills/improve-loop/SKILL.md- Versions recorded
- 3
- Latest version
- git:20260930.c1b179f, committed Sep 30, 2026
- First version recorded
- Sep 2, 2026
- Size
- 2,225 bytes
- Repository stars
- 95
- License
- none declared
Run an autoresearch loop that hill climbs on a project's evaluation score. Edit only allowed files, one change at a time, rerun the evaluation, keep a change only when the score improves, and log every attempt until the budget is spent. Use when a project provides a program.md that names the editable files, the evaluation command, and the budget.
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Improve loop An autoresearch loop for any project. The human writes `program.md`; you run the loop it describes. ## Setup 1. Find the project's `program.md` (e.g., `optimize/program.md`) and read it fully. It defines: - the files you may edit, - the files you must never edit (the evaluator, the tests, the held-out cases, and the budget state), - the evaluation command and the metric it reports, - the keep rule, the budget, and the log file. 2. If any of the above is missing, stop and ask the human. Never guess an evaluation command or a metric. 3. Run the evaluation once on the current code to record a baseline score, and log it as the current best. ## Loop 1. Read the latest evaluation result and pick one failure to fix. 2. Make one small change in the allowed files only. Prefer deleting or simplifying code over adding more. 3. Commit the change, then run the evaluation command. 4. Keep the commit only if the keep rule in `program.md` is met. Otherwise revert the edited files to the last kept commit. 5. Append one line to the log: the change, the files changed, a one-line rationale, the score, the decision, the commit, and the result file. …
Read the whole file at its exact version.
How to install
mdr add ai-evals-course/cartwheel-homeworks/improve-loop@sha256:2d3a122b2ed0d739The line at the top of this page pins the label git:20260930.c1b179f, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_jgbmlqxzibpbq5p2)
3 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 3 versions of this file, between Sep 2, 2026 and Sep 30, 2026, about one every 14 days.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20260930.c1b179f latest | 2026-09-30 | c1b179f | 2,225 B | A | view · diff |
| git:20260930.2acad7b | 2026-09-30 | 2acad7b | 2,075 B | A | view · diff |
| git:20260902.cdf1964 | 2026-09-02 | cdf1964 | 1,760 B | A | view |
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (2225 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
ai-evals-course/cartwheel-homeworks · 95 stars · license none declared · pushed 2026-10-10 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_jgbmlqxzibpbq5p2 GET https://markdownregistry.com/api/v1/resolve?ref=ai-evals-course/cartwheel-homeworks/improve-loop GET https://markdownregistry.com/api/v1/blob/2d3a122b2ed0d7390d4bc3486fc2cec8c4c0780ed8d4fa84a9a0801b388bf49f
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.
More from ai-evals-course/cartwheel-homeworks
Every file in ai-evals-course/cartwheel-homeworks