product-brain-starter
AGENTS.md1 starsChanged 4 days agoMITView on GitHub
Fill in every section marked FILL IN. Keep this file short enough to read in two minutes. It is loaded on every turn, and the more instructions a model is given at once, the more it misses. Only rules with no exceptions by area belong here. Area rules go in brain/rules/.
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 33,315 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 repository5 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- brain-healthSkill
- captureSkill
- plan-in-nine-stepsSkill
- write-handoffSkill
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
- Very specific: line 22
Only rules with no exceptions by area belong here.
- Very specific: line 32
Treat status: proposed entries and anything in brain/inbox/ as undecided.
- Distinctive: line 35
Ask rather than assume. If a question blocks the work, stop and ask.
- Very specific: line 36
Never set status: approved yourself.
- Distinctive: line 38
If an entry looks out of date or conflicts with another, say so.
≈ “Common” is an estimate: at least 1 in 5 of the 42,283 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
Product brain: always-loaded instructions
Fill in every section marked FILL IN. Keep this file short enough to read in two minutes. It is loaded on every turn, and the more instructions a model is given at once, the more it misses.
The product in five lines
- FILL IN: what the product is. Example: A billing portal where small businesses pay invoices and request refunds.
- FILL IN: who uses it. Example: Account owners, finance staff and our support team.
- FILL IN: what matters most this quarter. Example: Fewer refund disputes.
- FILL IN: the main systems it touches. Example: Payments provider, invoice service, customer email.
- FILL IN: where the code lives. Example: github.com/your-team/billing-portal.
Glossary
| Term | Meaning |
| FILL IN | Example: "Wholesale" means a customer on a trade agreement with volume pricing. |
| FILL IN | Example: "Credit note" means a document that reduces an invoice after issue. |
Must-follow rules that apply everywhere
Only rules with no exceptions by area belong here. Area rules go in brain/rules/.
- FILL IN. Example: Never show full card numbers in any screen, log or message.
- FILL IN. Example: Amounts are stored in cents and shown in the customer's currency.
- FILL IN. Example: Anything that moves money needs an audit log entry.
Protocol
- Read
brain/INDEX.mdfirst. - Open only entries with
status: approvedwhose scope matches the area you are working on. Do not open the rest. - Treat
status: proposedentries and anything inbrain/inbox/as undecided. Never present them as fact. - Cite entry IDs (for example D-0001, R-0001) wherever you rely on them.
- Label every assumption as an assumption, with your confidence and what would disprove it.
- Ask rather than assume. If a question blocks the work, stop and ask.
- Never set
status: approvedyourself. Only a named person does that. Write proposals tobrain/inbox/. - Open files in
sources/only to check a claim an entry makes. Never load them whole. - If an entry looks out of date or conflicts with another, say so. Do not pick one quietly.
- To plan a feature, follow the nine planning steps in
steps/, one at a time, starting with the executive summary.
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.

