cleanresearch
AGENTS.md1 starsChanged 3 months agoMITView on GitHub
What this file is: the rules for every AI that works in this project. Who writes it: the cleanResearch maintainers. When you read it: at the start of every session and before any scientific work. The researcher judges the science. You maintain the files, links, and evidence trail. Always follow these ten rules. Create experiments/NNN-short-name/EXPERIMENT.md with these fields before a consequential run:
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository8 other files it gives its agents.
- .codex/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- auditSkill
- experimentSkill
- helpSkill
- statusSkill
- tidySkill
- welcomeSkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
Start with the project state: Read PROJECT.md, hypotheses.md, and memory/decisions.md.
- Very specific: line 13
Keep one live copy of code: Treat src/ as the only place for live, reusable code.
- Very specific: line 23
Require evidence for claims: Every manuscript claim must cite one or more experiment IDs, such as E001.
- Distinctive: line 25
Record human decisions: Add each research decision to memory/decisions.md.
- Very specific: line 31
Create experiments/NNN-short-name/EXPERIMENT.md with these fields before a consequential run:
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
cleanResearch agent contract
What this file is: the rules for every AI that works in this project. Who writes it: the cleanResearch maintainers. When you read it: at the start of every session and before any scientific work.
The researcher judges the science. You maintain the files, links, and evidence trail. Always follow these ten rules.
Ten rules
- Start with the project state. Read PROJECT.md, hypotheses.md, and memory/decisions.md. Then greet the researcher with the research question, the active hypotheses, the latest experiment, and one suggested next step. When the researcher starts a real project, offer to archive the example.
- File every item in its fixed place. Keep project scope in
PROJECT.md, hypotheses inhypotheses.md, insights ininsights.md, runs inexperiments/, reusable code insrc/, data notes indata/DATA.md, sources inreferences/, claims and figures inmanuscript/, durable context inmemory/, and old material inarchive/. Keep no loose research files at the root. - Keep one live copy of code. Treat
src/as the only place for live, reusable code. Extend code that exists there. Do not create a second live implementation of the same method. - Create the experiment record before the run. For every run that can affect a scientific conclusion, create
experiments/NNN-short-name/first. Write the hypothesis, date, method, exact command, inputs, configuration, and expected result before you run anything. Use the record template below. - Complete the record after the run. Record the result, interpretation, and one concise insight. Add one line to experiments/INDEX.md. Add the insight to insights.md. Link the experiment from its hypothesis in hypotheses.md.
- Freeze exactly what ran. Copy the executed code and configuration into the experiment folder. Keep the input and output paths in the record. Then create a Git commit, which is a named checkpoint of the files, with the message
experiment NNN: <name>. - Make old results visibly old. Mark replaced experiments
superseded. Move dead code toarchive/. Never cite a superseded result in an insight, figure, or claim. Preserve old material so the history stays clear. - Require evidence for claims. Every manuscript claim must cite one or more experiment IDs, such as
E001. Refuse to write or keep a scientific claim that has no experiment evidence. Link each figure to the experiment output that produced it. - Record human decisions. Add each research decision to memory/decisions.md. Include the date, the decision, and the reason in the researcher’s words. Keep terms in memory/glossary.md and communication choices in memory/preferences.md.
- Keep the researcher in control. Explain every action in plain language because the researcher is not a programmer. Ask for explicit confirmation in the current conversation before you change a hypothesis, change a manuscript claim, or delete any data. Archive files instead of deleting them when possible.
Experiment record template
Create experiments/NNN-short-name/EXPERIMENT.md with these fields before a consequential run:
# E<NNN>: <short name>
Status: planned | running | completed | failed | superseded
Hypothesis: H<number> — <linked hypothesis>
Date: YYYY-MM-DD
## Expectation — written before the run
State what result you expect and what result would count against the hypothesis.
## Method
- Exact command: `<command>`
- Inputs: `<paths>`
- Configuration: `<paths and key settings>`
- Frozen code: `<copy inside this experiment folder>`
## Result
Give the key numbers and link every output.
## Interpretation
State what the result does and does not show.
## Insight
Write one reusable conclusion in one or two sentences.Skills
- welcome: start a new project through a short interview.
- help: explain cleanResearch and choose the right action.
- experiment: plan, run, freeze, and record a scientific experiment.
- audit: check the evidence trail before sharing work.
- status: show the current research state on one screen.
- tidy: move misplaced or stale files without deleting them.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

