darwin / knowledge
AGENTS.md0 starsChanged 2 months agoMITView on GitHub
One note per file. INDEX.md is the entire retrieval mechanism. INDEX.md is in context already. If a line matches the task, open that one file. If no line matches, move on. There is no second search step and no fallback scan of cards/. Write a note when something cost real effort to work out and would cost it again: a pitfall, a verified workaround, a fact the model would not already know, or a correction from the user. Write it…
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 31,439 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 repositoryOne other file it gives its agents.
- darwinSkill
The file
We split this AGENTS.md into the rules it gives an agent
6 rules. About 1 is common, 4 are distinctive, and 1 is very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Write a note when something cost real effort to work out and would cost it again: a pitfall, a verified workaround, a fact the model would …
- Distinctive: line 14
Write it when you find it rather than waiting for it to happen a second time.
- Distinctive: line 17
If an existing card covers the domain or topic, extend or refine that card rather than creating a fragmented duplicate.
- Distinctive: line 20
Do not write a note for anything the project itself records (code structure, git history, project conventions), anything that only matters …
- Very specific: line 67
Use --user when the change was explicitly commanded by the human user.
- Common: line 71
No credentials, no API keys, no tokens.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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
Knowledge Base Rules
One note per file. INDEX.md is the entire retrieval mechanism.
Reading
INDEX.md is in context already. If a line matches the task, open that one file. If no line
matches, move on. There is no second search step and no fallback scan of cards/.
Writing
Write a note when something cost real effort to work out and would cost it again: a pitfall, a verified workaround, a fact the model would not already know, or a correction from the user. Write it when you find it rather than waiting for it to happen a second time.
Before creating a new card, grep -i <topic> INDEX.md:
If an existing card covers the domain or topic, extend or refine that card rather than creating
a fragmented duplicate.
Do not write a note for anything the project itself records (code structure, git history, project conventions), anything that only matters inside the current conversation, anything the model already knows, or anything you have not verified.
Capture is one file plus one index line:
cards/<slug>.md frontmatter: title, last_verified, confidence (or created/updated)
INDEX.md one line carrying the symptom, not just the topicBoth belong in the same commit. Split across two commits, reverting the note leaves an index line pointing at a file that no longer exists.
Correcting & Retracting
Find out when the wrong claim entered the file before deciding what to do about it:
git log -S'<the wrong sentence, verbatim>' --oneline -- cards/<slug>.mdPickaxe search (-S) finds the commit that introduced the text, not the one that last touched the
line. Three cases follow, and they need different responses:
- Stale. It was right and the environment moved. Add a version or date boundary (
scope) and keep the content. - Misjudged. It was wrong when written. Apply a paragraph-level retraction (
retract): rewrite that paragraph as "previously claimed X, which does not hold, because Y". Keep the file, keep the index line, keep the symptom description. The symptom is the retrieval key; deleting the note removes the only thing that would catch the same problem next time, causing future agents to research and write the same wrong note again. - Broken by an edit. A later change turned a correct note wrong. Revert that change (
revert) instead of editing on top of it.
Symptoms are observations and are usually right. Root causes are inferences and are what turn out wrong. Fixes sometimes keep working even when the root cause is wrong, which is how a bad diagnosis survives. A whole note is rarely wrong, so default to correcting the root cause alone.
Logging
Log changes via darwin.sh:
darwin.sh [--user] <action> <file> [more-files...] <one-line-why> < /dev/nullUse --user when the change was explicitly commanded by the human user.
Never
No credentials, no API keys, no tokens. Take care with identifiers too: device IDs, internal hostnames, account IDs, zone IDs, and private paths are what keeps a knowledge base private.
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.

