project-starter-template-en-ai
AGENTS.md2 starsChanged 2 months agoMITView on GitHub
Instructions for the AI coding agents working in this repository. [One or two lines on what this project is and does.] The full documentation lives in docs/ — this file is its only index; read it before any non-trivial work. architecture/ vs conventions/: architecture/ answers what this project builds (its data model, its API, its screens); conventions/ answers how the work is done (how it is tested, how it is deployed, how secrets are handled). Both are filled in when…
- Reads credentials
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 repository20 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- accessibility-auditSkill
- changelogSkill
- commitSkill
- configure-repoSkill
- copywritingSkill
- design-system-auditSkill
- i18n-paritySkill
- identitySkill
- instantiateSkill
- migration-guardSkill
- new-adrSkill
- new-specSkill
- open-issueSkill
- open-prSkill
- port-changeSkill
- prototypeSkill
- releaseSkill
- seo-auditSkill
- update-templateSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
13 rules. About 0 are common, 4 are distinctive, and 9 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 63
Accompany actively: Ask when scope or priority is ambiguous instead of assuming.
- Very specific: line 67
No spec, no change (mandatory): Every feature or fix (feat/*, fix/*) is born from a spec in specs/ sharing the branch's slug (create it …
- Very specific: line 81
Issues and PRs from the templates: When creating an issue or a pull request, ALWAYS use the templates in .github/ (ISSUE_TEMPLATE/ and …
- Very specific: line 86
Tests with every change: Follow docs/conventions/testing.md.
- Very specific: line 87
"Done" has a definition: Before calling a change done, go through docs/conventions/definition-of-done.md.
- Very specific: line 89
Keep the documentation in sync: Update the relevant docs/ and CHANGELOG.md.
- Very specific: line 94
Never leave a change undocumented: every change updates CHANGELOG.md and the affected docs/ in the same PR, and every implemented spec …
- Very specific: line 99
Never commit secrets or real .env values.
- Distinctive: line 101
Never invent dependencies, files or APIs — verify they exist first.
- Very specific: line 102
Never edit code on a feat/*/fix/* branch without its spec in specs/, nor create work branches off main — the hooks block those two rules.
- Distinctive: line 104
Never bypass authorization checks or weaken security to make something work.
- Distinctive: line 105
Never push to main or force-push to shared branches.
- Distinctive: line 106
Do not reformat unrelated code or make sweeping changes outside the task.
≈ “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
<!--
Canonical instructions for [PROJECT_NAME]'s agents.
This is the single source of truth for AI coding agents (Claude Code, Copilot,
Cursor, etc.). CLAUDE.md imports this file. Keep it concise (<150 lines) and
include only non-obvious, project-specific guidance — people read the README.
-->AGENTS.md — [PROJECT_NAME]
Instructions for the AI coding agents working in this repository.
Project summary
[One or two lines on what this project is and does.] The full documentation lives in
docs/ — this file is its only index; read it before any non-trivial work.
Repository map (read this first)
| You need to know… | Read |
| How to work here without reviewing code | docs/conventions/ai-agents.md |
| The work cycle, from spec to release | docs/conventions/workflow.md |
| Why the product exists and how it is paid | docs/product/business-model.md |
| What is done and what comes next | docs/product/roadmap.md |
| Which stack was chosen here | docs/architecture/stack.md |
| How the system is built | docs/architecture/architecture.md |
| Data model | docs/architecture/database.md |
| Authentication and permissions | docs/architecture/auth.md |
| The API contract | docs/architecture/api.md |
| Which screens exist and where each one is | docs/architecture/screens.md |
| How we write code | docs/conventions/ |
| Why each decision was taken | docs/decisions/ |
| Visual identity and tokens | design/ |
| The design system, for AI agents | DESIGN.md |
architecture/vsconventions/:architecture/answers what this project builds (its data model, its API, its screens);conventions/answers how the work is done (how it is tested, how it is deployed, how secrets are handled). Both are filled in when instantiating — the difference is not "reusable vs own" but the question they answer. When a topic does not warrant both questions, it lives in a single document (auth lives entirely inarchitecture/auth.md, rules included); only the database keeps the pair, and each rule lives on exactly one side.
Why
design/is not insidedocs/:docs/is what you read to build;design/is consumed —design/tokens.cssis imported by the application's CSS. When the project has an application, its content will migrate inside it (app/assets/) and only its guide will stay at the root.
Always defer to
docs/conventions/for style and rules — they are the source of truth, not your priors.
Setup & commands
[INSTALL_DEPENDENCIES_COMMAND] # install dependencies
[START_DEV_COMMAND] # run locally
[TEST_COMMAND] # run the test suite
[LINT_COMMAND] # lint / formatWorking agreement
- Accompany actively. Ask when scope or priority is ambiguous instead of assuming;
suggest what the person did not ask for but will need. Priority is set by
docs/product/roadmap.md. The full cycle is indocs/conventions/workflow.md. - No spec, no change (mandatory). Every feature or fix (
feat/*,fix/*) is born from a spec inspecs/sharing the branch's slug (create it with/new-specor delegate to thearchitectsubagent), and itsproposal.mddeclares which roadmap item it completes. Thespec-guardrails.shhook guarantees it: with no spec —or with a half-filledproposal.md— you cannot edit code. - Follow Git Flow. Work branches (
feat/…,fix/…,docs/…,chore/…) are ALWAYS born offdevelop, never offmain(onlyhotfix/*starts frommain). Ifdevelopdoes not exist, create it frommainand publish it before any work. SeeCONTRIBUTING.md. Never commit directly tomainordevelop. Thegit-guardrails.shhook blocks the violations. - Code is written in English. Variables, functions, classes, files, tables, columns and paths go in English, regardless of the project's or the conversation's language.
- Conventional Commits.
type(scope): summary. Add the AI co-author line for AI-assisted commits (seedocs/conventions/ai-agents.md). - Issues and PRs from the templates. When creating an issue or a pull request, ALWAYS
use the templates in
.github/(ISSUE_TEMPLATE/andPULL_REQUEST_TEMPLATE.md) and respect the branching inCONTRIBUTING.md. The/open-issueand/open-prskills do it for you. - Tests with every change. Follow
docs/conventions/testing.md. - "Done" has a definition. Before calling a change done, go through
docs/conventions/definition-of-done.md. - Keep the documentation in sync. Update the relevant
docs/andCHANGELOG.md; record notable decisions as an ADR indocs/decisions/.
Strict rules — never do this
- Never leave a change undocumented: every change updates
CHANGELOG.mdand the affecteddocs/in the same PR, and every implemented spec ticks its item indocs/product/roadmap.md; every notable decision leaves an ADR. A change with no documentation is not done — thechangelogjob inquality.ymlverifies it on every PR. - Never commit secrets or real
.envvalues. SeeSECURITY.mdanddocs/conventions/secrets.md. - Never invent dependencies, files or APIs — verify they exist first.
- Never edit code on a
feat/*/fix/*branch without its spec inspecs/, nor create work branches offmain— the hooks block those two rules. - Never bypass authorization checks or weaken security to make something work.
- Never push to
mainor force-push to shared branches. - Do not reformat unrelated code or make sweeping changes outside the task.
AI assistants and tooling
- This file is the canonical context and the single documentation index. Tool-specific
files (
CLAUDE.md, etc.) point here. - The subagents live in
.claude/agents/and the skills in.claude/skills/. Interviews and anything that requires asking the person something go in a skill: a subagent runs in its own context and has no way to ask anything. - Working method, where to verify and where you fail silently:
docs/conventions/ai-agents.md.
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.

