Dev-Labs
AGENTS.md68 starsChanged 18 days agoMITView on GitHub
Before every source push, increase the app's Major.Minor.Patch version with npm run version:patch, npm run version:minor, or npm run version:major. Commit package.json and package-lock.json with the pushed work. Use minor for a new backwards-compatible feature, major for a breaking change, and patch for fixes and routine updates. See RELEASE.md. Orientation for AI coding agents (Claude Code, Codex, Cursor, …) and human contributors. Read this first, then docs/ARCHITECTURE.md for the deep map. DevLabs is an open-source (MIT) Electron desktop app, forked…
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 repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ian-xiaohei-illustrationsSkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 3
Before every source push, increase the app's Major.Minor.Patch version with npm run version:patch, npm run version:minor, or npm run …
- Very specific: line 5
Commit package.json and package-lock.json with the pushed work.
- Distinctive: line 10
Read this first, then docs/ARCHITECTURE.md for the deep map.
- Very specific: line 45
i18n parity: Every UI string is in en, zh-CN and ar, with the same keys and interpolation variables.
- Very specific: line 64
After code changes, run graphify update .
≈ “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
AGENTS.md — working in this repo
Before every source push, increase the app's Major.Minor.Patch version with
npm run version:patch, npm run version:minor, or npm run version:major.
Commit package.json and package-lock.json with the pushed work. Use minor
for a new backwards-compatible feature, major for a breaking change, and patch
for fixes and routine updates. See RELEASE.md.
Orientation for AI coding agents (Claude Code, Codex, Cursor, …) and human contributors. Read this first, then docs/ARCHITECTURE.md for the deep map.
What this is
DevLabs is an open-source (MIT) Electron desktop app, forked from Munder Difflin by Chaitanya Giri. It runs a team of terminal AI agents (Claude Code, Codex, Antigravity, Cursor and others) in a pixel-art office, with an orchestrator named Ramon.
There is no account, sign-in or license check. Agents start as soon as the app is installed.
Layout
| Path | What |
src/main/ | Electron main process: PTYs (pty.ts), hive/agent coordination (hive.ts), MCP proxy (integrationBroker.ts), hires/floors I/O (hire.ts), IPC (index.ts) |
src/preload/index.ts | The only renderer↔main bridge (window.cth), fully typed |
src/renderer/src/ | React 18 + zustand UI; office scene (Pixi.js); components/; i18n/locales/{en,zh-CN,ar}.json |
src/shared/ | Pure, dependency-light modules shared by main and renderer: validators (hire.ts, floor.ts), MCP catalog (mcpCatalog.ts) brand (brand.ts) |
test/*.test.cjs | App tests (node --test, TS loaded via test/load-ts.cjs) |
docs/, .documentations/ | Contributor docs / user- and owner-facing guides |
Commands
npm ci # app deps (needs a C/C++ toolchain for node-pty, better-sqlite3)
npm run typecheck # main + renderer
node --test test/*.test.cjs # app tests (= npm run test:focused)
npm run build # electron-vite build
npm run dev # run the appCI (.github/workflows/) runs: app typecheck, check:links and build.
test/proc-kill.test.cjs fails in sandboxed Linux containers; that's a known issue that predates this work.
Rules that are enforced by tests. Don't break them.
- i18n parity. Every UI string is in
en,zh-CNandar, with the same keys and interpolation variables. - Shared manifests are untrusted. Hires and floors go through
validateHireManifest/validateFloorManifest:- they never carry a binary or a raw command (the command is built locally from the provider preset);
- flags must come from the safe list;
- MCP/skill ids must come from the allowlists;
- importing never spawns.
- Connector keys live only in Electron
safeStorage. HTTP connectors reach the vendor through the loopback broker, so the agent never sees the key. Keyed catalog entries are never on by default. - MCP servers go in
--mcp-config, never--settings. Claude Code ignores the latter.
Conventions
- Match the surrounding code: comment density, naming, inline styles with
--cth-*tokens (seeDESIGN.md). - Renderer ↔ main only through
window.cth. Add the IPC handler insrc/main, the method insrc/preload/index.ts, and a test. - Pure logic goes in
src/shared/so it's testable withloadTsand no Electron. - User-facing text: plain words, no jargon; errors say what to do next.
- Docs to update with a feature:
CHANGELOG.md(Unreleased);- the relevant
.documentations/*.mdordocs/*.md; docs/USER_GUIDE.mdif customers see it.
- After code changes, run
graphify update .if graphify is installed (seeCLAUDE.md).
Where things are documented
| Topic | Doc |
| Customer guide | docs/USER_GUIDE.md |
| Floors / connectors | .documentations/FLOORS.md, .documentations/CONNECTORS.md |
| Legal | LICENSE (MIT), PRIVACY.md, THIRD_PARTY_NOTICES.md, LICENSE-ASSETS, CONTRIBUTORS.md |
| Telemetry contract | TELEMETRY.md (the code's allowlist must match it) |
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.

