One folder, every AI agent recognizes it.
Drop one folder. Any AI agent picks it up.
The standard tells Claude Code, Cursor, Windsurf — or your own homebrew agent — where the project's tasks live, where context goes, where decisions get logged, and which rules to follow. No setup beyond pasting one prompt.
Want to evaluate first? View the prompt as plain text before pasting it into your agent.
Read https://meshkore.com/standard.md and apply it to this repo.
Your agent fetches the spec, creates .meshkore/, drops the editor rules at the repo root, and reports back what changed — everything reviewable in the diff.
.meshkore/ in your repo. Offline. No account.What you'll get when you adopt this
The agent creates a single folder — .meshkore/ — in your repo. That folder becomes the home for everything an AI-driven project needs. You won't have to invent any of this yourself, and any AI agent that joins the project later will recognize it instantly.
Every initiative, task, and cycle written down in plain markdown your agent reads — instead of asking what to do next.
Architecture notes, conventions, gotchas, decisions. Every new AI session starts informed instead of from zero.
A dedicated folder for tokens and secrets — gitignored, never streamed into prompts, easy to audit.
The build, deploy, audit, recovery playbooks your agents run daily — versioned, reusable across projects.
A CLAUDE.md / .cursorrules / .windsurfrules at the repo root so every new session loads the conventions on its own.
The same folder becomes the home for multiple agents on multiple devices — coordinated from Architect UI when you're ready.
Standard vs Cluster — what's what
Two related concepts that get conflated. The standard is the folder convention; a cluster is what you get when you turn the daemon on and connect several machines. You always start with the standard; a cluster is optional, layered on top.
Drop .meshkore/ in any repo and your AI agent immediately knows where to find context, where to write tasks, how to log decisions, and which rules to follow. Works offline. No account. Zero infrastructure.
Affects: one repo. Read by: every AI editor (Claude Code, Cursor, Windsurf, …).
Install the daemon (meshcore) and your project becomes a cluster — multiple machines and AI agents working in parallel on the same repo, coordinated, observable from Architect UI. Optional. You climb here when you need it.
Affects: one project across N machines. Read by: the daemon + Architect UI + (optionally) the agent network.
Why this exists
Every AI session today starts from zero: you re-explain the codebase, re-state conventions, re-list what not to touch. Multiply that across editors, machines, and team members, and you spend more time briefing the agent than building. Worse, conventions drift — what's in your head, your README, and your .cursorrules diverge.
The MeshKore Standard fixes this with a single, opinionated layout under .meshkore/. Agents that follow it pick up project context the same way every time. You stop re-briefing; you start shipping.
How it works, in plain language
A repo "applies the standard" when it has a .meshkore/ folder with a predictable shape: docs in one place, tasks in another, governance in a third, secrets fenced off. Add an editor rules file at the repo root (CLAUDE.md, .cursorrules or .windsurfrules) so every new agent session loads the conventions automatically. That's the whole minimum.
Everything else — registering your agent on the public hub, running our local daemon for automation, joining the cross-cluster mesh — is opt-in. You climb the ladder when you need a capability, not before.
The five layers
.meshkore/ tree with docs, tasks, governance. Works offline, no account, no install. The starting point for almost everyone.
CLAUDE.md / .cursorrules / .windsurfrules at the repo root so every agent session loads the conventions on its own.
meshcore locally for automation, a live web visor of your cluster, and headless dispatch to your AI clients.
Each row is a strict superset of the one above. Stop wherever your problem stops. Full decision tree + per-layer setup at /cluster/adopt.
Where things live
The short version of the folder layout:
.meshkore/ ├── public/cluster.yaml ← only committed to git ├── docs/ project context ├── modules/<id>/ tasks/log/diagrams ├── log/<YYYY-MM-DD>.md daily prose ├── timeline/<date>.jsonl machine events ├── credentials/ secrets, never committed └── (a few more) full list in spec
Cycles — optional timebox (v11+)
Tasks gain one optional frontmatter key: cycle: <id>. A cycle is a 1–2 week timebox (Linear's primitive) declared by a file in .meshkore/cycles/<id>.md:
--- id: 2026-w22 title: "Sprint 22 — CC P0" starts: 2026-05-25 ends: 2026-06-07 status: active goal: "Close CC01..CC06" ---
Operators who don't want sprint cadence simply don't create cycle files — every task stays uncycled and nothing changes.
Versioning
The standard is at v11. Breaking changes get a major bump and an explicit migration path.
The changelog describes every version; the LLM applies the catch-up when a cluster's .meshkore/STANDARD_VERSION falls behind.
Operators who follow adoption ladder never see migrations — the daemon detects drift and prompts.
For agents and automation
This page is the human overview. The full normative content — schemas, frontmatter, governance R1-R6, path conventions, hard rules, editor boot block — lives in the agent-facing spec. If you're an AI agent applying the standard to a repo, fetch one of these instead:
Refresh cadence: once per session, or every 24h. The updated field in standard.json signals changes. If anything elsewhere on this site contradicts that spec, the spec wins.
Start here
Drift between this page and any other? File it as a doc-bug at github.com/meshkore. The agent-facing spec is the tiebreaker.