hivemind
AGENTS.md17 starsChanged 12 days agoMITView on GitHub
Guidance for AI agents (and humans) working in this repo. Subfolders have their own AGENTS.md with local detail — read the one nearest the code you're changing. For the research/knowledge-base workflow, see CLAUDE.md. A Linux desktop app (Electron + an infinite xyflow canvas) that is mission-control for AI coding agents. Every tile is a live terminal / diff / file-tree / editor / issues board; tiles live in frames (workspaces) bound to a real repo on disk — local, a…
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 repository11 other files it gives its agents.
- guide/AGENTS.mdAGENTS.md
- spec/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- adding-an-agent-providerSkill
- ascii-character-animationSkill
- feature-doneSkill
- hive-browserSkill
- hivemind-custom-viewSkill
- hivemind-live-wallpaperSkill
- maintaining-docsSkill
- test-auditSkill
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
- could be automatic: a setting could enforce it
- Very specific: line 43
Run the relevant e2e when you touch canvas/frame/tile/issue behavior.
- Very specific: line 52
Design tokens: use the var(--color-*) palette + the .u-eyebrow utility.
- Very specific: line 56
Commits: no Co-Authored-By or Claude trailers.
- Very specific: line 68
Never build/release locally.
- Distinctive: line 85
No previous sessions found.
≈ “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
AGENTS.md — hivemind
Guidance for AI agents (and humans) working in this repo. Subfolders have their
own AGENTS.md with local detail — read the one nearest the code you're
changing. For the research/knowledge-base workflow, see CLAUDE.md.
What this is
A Linux desktop app (Electron + an infinite xyflow
canvas) that is mission-control for AI coding agents. Every tile is a live
terminal / diff / file-tree / editor / issues board; tiles live in frames
(workspaces) bound to a real repo on disk — local, a git worktree, or a remote
SSH host. Issues are plain markdown under .hivemind/; agents read/update them
with the hive CLI (hive ctl is the control plane). Local-first, no cloud, no telemetry.
Monorepo layout
apps/
desktop/ Electron main + preload + React renderer (the canvas app) → apps/desktop/AGENTS.md
cli/ `hive` CLI (citty + bun-compile); `hive ctl` = agent control plane (HCP client)
packages/
hive-core/ storage + parsing for .hivemind/ (gray-matter + zod), skill templates + agentic installer → packages/hive-core/AGENTS.md
hive-agents/ the agent-provider catalog: one def (+ node half) per CLI agent, read by UI/CLI/HCP
tsconfig/ shared TS config
templates/ source of the hive-browser skill (embedded into hive-core)
docs/design/ architecture design docs (e.g. remote-frames.md)
scripts/ release.sh + helperspnpm workspace; Node ≥ 22, pnpm 12 (packageManager in package.json pins the exact version), bun ≥ 1.1 (CLI compile only).
Build / test / verify (run from apps/desktop unless noted)
pnpm run typecheck # tsc --noEmit, web + node projects — ALWAYS run before commit
pnpm run build # electron-vite build (validates the real bundle path)
pnpm test:unit # node:test — fast, pure logic; add tests here first
pnpm test:e2e # Playwright + xvfb — needs `unset ELECTRON_RUN_AS_NODE` firstThe minimum gate before any commit: typecheck + build + test:unit green. Run the relevant e2e when you touch canvas/frame/tile/issue behavior.
Conventions
- TypeScript strict. No
anyescapes without reason. - Comments document why, especially load-bearing trade-offs and the non-obvious. Match the density of the surrounding file. No restating the code.
- Icons: from
components/iconsonly — never emoji or unicode glyphs as UI icons (they render inconsistently and aren't theme-colorable). - Design tokens: use the
var(--color-*)palette + the.u-eyebrowutility; don't hard-code hex. Informational text uses--color-fg2(passes WCAG AA);--color-fg3is decoration only. Addaria-labelto icon-only buttons and a focus ring to every input. - Commits: no Co-Authored-By or Claude trailers. Work on a branch and open a PR;
don't commit to
main. Commit in logical chunks. Keep stray files (.rig/, scratch) out of commits — add explicit paths, don'tgit add -Ablindly. - CHANGELOG hand-off rule: anything that ships to users gets a one-line entry
under
## [Unreleased]inCHANGELOG.mdbefore you hand back.
Writing docs
Writing or editing docs in guide/ — follow .claude/skills/maintaining-docs/.
Releases
Never build/release locally. ./scripts/release.sh (calendar version, computed) from a
clean main bumps versions, writes the changelog section, tags, and pushes —
GitHub Actions builds + publishes. Do not run it unless explicitly asked.
Gotchas
- e2e:
unset ELECTRON_RUN_AS_NODEbeforeplaywright, or Electron starts as plain node. - A persistent PTY daemon survives the window; remote PTYs run in-main (they
can't survive an ssh drop). See
apps/desktop/AGENTS.md.
<claude-mem-context>
# Memory Context[hivemind] recent context, 2026-09-08 12:41pm GMT+5:30
No previous sessions found.
</claude-mem-context>
Website hosting
The website lives in site/, and its guide's pages in guide/, built by the Fumadocs app in
site/docs/ and served at /guide. .github/workflows/site.yml deploys both to Cloud Run when a
change under site/ or guide/ reaches main. docs/ holds engineering notes (design, plans,
reviews) and is not published. Do not create or deploy ChatGPT Sites for this project.
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.

