Back to search
agents-hub
MCP serverGitHub
One source of truth for AI agent skills, subagents, MCP servers and global instructions (Claude Code, Codex, Cursor)
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/sujit-gholap-121--agents-hub)HTML
<a href="/listings/sujit-gholap-121--agents-hub"><img src="/api/listings/sujit-gholap-121--agents-hub/badge.svg" alt="findsafeskills safety rating" /></a>Install
i
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 95 lines
- # agents-hub
- The single source of truth for my AI agent setup. Claude Code, Codex and Cursor all get their
- **skills, subagents, MCP servers and global instructions** from here, via one command:
- ```sh
- agents-sync
- ```
- Backed up to the public repo `sujit-gholap-121/agents-hub` (every sync commits and pushes).
- ## Layout
- ```
- ~/.agents/
- ├── skills/<name>/SKILL.md skills (flat, the open Agent Skills format)
- ├── agents/<name>.md subagents (Claude-style markdown + frontmatter)
- ├── mcp/servers.json MCP servers, with ${VAR} placeholders for secrets
- ├── instructions/AGENTS.md global rules for every agent
- ├── experimental/{skills,agents} only switched on with --with-experimental
- ├── groups.json which group each skill belongs to
- ├── groups/ generated browsing view (links), don't edit
- ├── bin/agents-sync the sync tool
- ├── .env secrets for servers.json (NOT in git, chmod 600)
- └── state/ what the tool manages, logs, config backups (NOT in git)
- ```
- ## What goes where
- | Hub | Claude Code | Codex | Cursor |
- |---|---|---|---|
- | skills | links in `~/.claude/skills` | reads `~/.agents/skills` directly | reads `~/.agents/skills` directly |
- | agents | links in `~/.claude/agents` | generated `~/.codex/agents/*.toml` | links in `~/.cursor/agents` (app) + per-project (CLI, see below) |
- | MCP | `claude mcp add-json -s user` | generated block in `~/.codex/config.toml` | `~/.cursor/mcp.json` |
- | instructions | `~/.claude/CLAUDE.md` (link) | `~/.codex/AGENTS.md` (link) | `~/.cursor/rules/agents-hub.mdc` (always-apply rule) |
- **Cursor CLI and agents:** `cursor-agent` only loads subagents from the project it is opened in.
- Register a project once and every sync keeps it current (its git ignores the links locally):
- ```sh
- agents-sync --project ~/Workspace/Github/escalara-portal
- agents-sync --unproject ~/Workspace/Github/escalara-portal
- ```
- ## How a sync works
- 1. **Import.** Anything new that an installer put straight into one app (`npx skills add`,
- the Agency Agents app, `claude mcp add`, a hand-made Codex agent) is moved into the hub.
- Secret-looking MCP values go to `.env` and become `${VAR}` placeholders.
- 2. **Normalize.** Agent descriptions over 160 characters are shortened so all of them fit Claude
- Code's 15k-token limit. The original text is kept in `full_description:`.
- 3. **Export.** The hub is written into each app. Only entries the tool manages are touched.
- App-owned items stay where they are: `~/.claude/skills/synced`, `~/.codex/skills/.system`,
- `~/.cursor/skills-cursor`, Codex plugin MCPs, and MCPs listed in `servers.json` → `unmanaged`.
- 4. **Group, commit, push.**
- ## Everyday use
- ```sh
- # new skill
- npx skills add <repo> # installs into ~/.agents/skills
- # or: mkdir ~/.agents/skills/my-skill && $EDITOR ~/.agents/skills/my-skill/SKILL.md
- # new agent
- $EDITOR ~/.agents/agents/engineering-my-agent.md # prefix = group
- # new MCP server: add to mcp/servers.json, e.g.
- # "github": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"],
- # "env": { "GITHUB_TOKEN": "${GITHUB_TOKEN}" }, "targets": ["claude", "codex", "cursor"] }
- # and put GITHUB_TOKEN=... in ~/.agents/.env
- # experiments
- mv my-skill ~/.agents/experimental/skills/ && agents-sync --with-experimental
- agents-sync # then restart the apps (they read these at startup)
- agents-sync --dry-run # show what would change, change nothing
- ```
- `targets` (default: all three) limits an MCP server to some apps. `"enabled": false` turns it off
- everywhere. `"extra": {"codex": {...}}` adds app-specific fields (for example Codex timeouts).
- ## New machine
- ```sh
- git clone https://github.com/sujit-gholap-121/agents-hub ~/.agents
- ln -s ~/.agents/bin/agents-sync ~/.local/bin/agents-sync
- # recreate ~/.agents/.env (secrets are never in git), then:
- agents-sync
- ```
- ## Undo
- Everything is in git: `git -C ~/.agents log`, then `git revert <commit>` and `agents-sync`.
- The pre-hub backups are `backup-skills-2026-11-07.tgz` and `backup-agents-2026-11-07.tgz`
- (not in git), and the last Codex config before each change is in `state/codex-config.toml.bak`.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.