aiworkspace / root-config
AGENTS.md19 starsChanged 3 months agoApache-2.0View on GitHub
This is the workspace root -- the parent directory containing all project repos. Open this directory in your editor. Everything follows nearest-wins: the closer a file is to the code you're changing, the higher its priority. When guidance conflicts, the closest file to the code wins. Before working in a project, check for context nearest to the code first: - Read existing code before editing. Match the style and patterns already in use. - Use the nearest config file (linter,…
- 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 repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- root-config/CLAUDE.mdCLAUDE.md
- autoresearchSkill
- production-gradeSkill
- tribunalSkill
The file
We split this AGENTS.md into the rules it gives an agent
8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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
- Distinctive: line 24
When guidance conflicts, the closest file to the code wins.
- Distinctive: line 28
Before working in a project, check for context nearest to the code first:
- Distinctive: line 43
Read existing code before editing.
- Distinctive: line 44
Use the nearest config file (linter, formatter, tsconfig) to the code you are changing.
- Distinctive: line 45
Run existing tests after changes.
- Very specific: line 67
All tool configs are mirrored for every developer — Cursor, Claude Code, Codex, and VS Code.
- Very specific: line 91
If you use HTTP Bearer MCP servers in Cursor or Claude Code, follow the env-loading step in <workspace-repo>/setup.md §4.1 …
- Very specific: line 94
Check tokens and Cursor/Claude Code HTTP hints: npm run mcp:check-secrets
≈ “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
Workspace
This is the workspace root -- the parent directory containing all project repos. Open this directory in your editor.
Projects
<!-- Add your projects here as you clone them alongside workspace/ -->| Project | Purpose |
workspace/ | AI agent skills, configs, and automation |
Resolution Order
Everything follows nearest-wins: the closer a file is to the code you're changing, the higher its priority.
| What | Workspace-wide (shared) | Per-project (override) |
| Instructions | This file (AGENTS.md at root) | <project>/AGENTS.md |
| Skills | .agents/skills/ | <project>/.agents/skills/ |
| Rules | .cursor/rules/ | <project>/.cursor/rules/ |
| Docs | docs/ repo (sibling) | <project>/docs/ |
When guidance conflicts, the closest file to the code wins.
Finding Context
Before working in a project, check for context nearest to the code first:
- Project-level:
<project>/AGENTS.md, README, docs, and any project-specific skills or config - Workspace-level: this file and shared skills/rules at the workspace root
Conventions
<!-- Replace these examples with your team's conventions -->- Branches:
feature/description,bug/description,hotfix/description - Commits: lowercase imperative (
add feature,fix bug,update config) - PRs: target default branch unless noted
Code Quality
- Read existing code before editing. Match the style and patterns already in use.
- Use the nearest config file (linter, formatter, tsconfig) to the code you are changing.
- Run existing tests after changes. Don't skip or weaken tests to make code pass.
- Handle errors explicitly. Don't swallow exceptions or ignore return values.
<!-- Add your team's baseline standards here (e.g., test coverage, logging, accessibility) -->MCP servers
MCP servers give agents shared tools. The workspace ships with:
context7— up-to-date documentation for libraries, frameworks, and tools (HTTP OAuth).
These are defined once and shared across tools, the same way AGENTS.md and skills are. The canonical config is .agents/mcp.json, written with bare ${VAR} placeholders — the syntax Claude Code resolves natively. Claude Code symlinks straight to it. Cursor requires ${env:VAR} instead (its own MCP client's requirement, incompatible with Claude Code's), and Codex (TOML) and VS Code (a different JSON schema) can't share the file at all — so Cursor, Codex, and VS Code all get derived twins regenerated from canonical on npm run sync.
.agents/mcp.json canonical config (JSON, bare ${VAR})
.mcp.json -> .agents/mcp.json Claude Code
.cursor/mcp.json Cursor (JSON twin, sync-derived, ${env:VAR})
.codex/config.toml Codex (TOML twin, sync-derived)
.vscode/mcp.json VS Code and Copilot (JSON twin, sync-derived)To add or change a server, edit .agents/mcp.json using bare ${VAR} for any secret placeholder. npm run sync refreshes the Cursor, Codex, and VS Code twins from canonical automatically.
All tool configs are mirrored for every developer — Cursor, Claude Code, Codex, and VS Code. There is no opt-out; unused configs are inert. To override workspace-wide MCP for one repo, use <project>/.cursor/mcp.json (nearest-wins).
npm run sync refreshes bundled servers from the aiworkspace template when they are missing from canonical, and preserves servers you added that are not in the template. If canonical already defines context7, sync leaves it unchanged (any transport or URL) — customize it there. For other bundled defaults, use per-project MCP (<project>/.cursor/mcp.json, nearest-wins).
On first use in a project, the tool asks you to approve the project MCP servers. Approve them to enable the tools.
Editor-specific team vs personal settings (Cursor, Claude, VS Code, Codex) live in root-config/README.md. Edit MCP servers only in .agents/mcp.json, then npm run sync.
MCP authentication
Secret-bearing MCP servers load tokens from .env.local at the parent workspace root.
| Server type | How secrets load |
Stdio with ${VAR} | npm run sync wraps with mcp-load-env.mjs — reads .env.local automatically |
| HTTP OAuth | Editor sign-in (Slack, Atlassian, etc.) — no .env.local |
| HTTP Bearer, Claude Code | canonical's bare ${VAR} resolves from the real process environment at connection time — no .env.local auto-load. Needs the one-time env load below just like Cursor. |
| HTTP Bearer, VS Code | twin gets envFile — reads .env.local directly |
| HTTP Bearer, Cursor | twin gets ${env:VAR}; resolves from the real process environment at startup, not envFile — needs the one-time env load below |
Setup (each developer, once):
cp .env.example .env.localat the parent workspace root- Fill in tokens (never commit
.env.local) - If you use HTTP Bearer MCP servers in Cursor or Claude Code, follow the env-loading step in
<workspace-repo>/setup.md§4.1 (<workspace-repo>= this repo folder name) - Restart your editor
Check tokens and Cursor/Claude Code HTTP hints: npm run mcp:check-secrets
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.

