One Playbook, Every Agent

Your agent configuration is scattered. Skills live in .claude/skills/, your MCP servers in .mcp.json, a slightly different copy in .cursor/mcp.json, instructions in AGENTS.md — and every new AI coding tool adds another folder. Keeping them consistent by hand is exactly the kind of chore agents were supposed to eliminate.

Today the AgentPlaybooks CLI closes that loop. agentplaybooks sync now generates the platform files missing from every enabled target, pull/push connect your local project to a hosted playbook, and the whole CLI doubles as a Claude Code plugin — so your agent can run the workflow for you.

Five platforms, one command

apb sync normalizes what it finds into the canonical agentplaybook.json manifest, then fills the gaps per target:

| Target | Skills | MCP servers | |---|---|---| | Claude Code / Cowork | .claude/skills/ | .mcp.json | | Cursor | .cursor/skills/ | .cursor/mcp.json | | ChatGPT / OpenAI Codex | .codex/skills/ | .codex/config.toml | | Google Antigravity | .agents/skills/ | — | | Nous Hermes Agent | ~/.hermes/skills/ | — |

Write a skill once in Claude Code, run apb sync --apply, and it shows up in Cursor, Codex, and Antigravity too — including your MCP server definitions, translated between JSON and Codex's TOML automatically.

A nice detail: Google Antigravity reads project skills from .agents/skills/, which is exactly AgentPlaybooks' portable store. Pull a playbook and it is Antigravity-ready with zero extra steps.

Safe by default

The sync engine keeps the guarantees from our original design:

• Plan first. Nothing is written or uploaded without an explicit

--apply.

• No silent overwrites. Same-named definitions with different content are

conflicts — reported and skipped until you resolve the drift.

• Backups. Any modified file is copied to .agentplaybooks/backups/

first.

• No secret leaks. Secret values never enter the manifest, and push

refuses to upload content that looks like it contains hard-coded credentials.

Team playbooks: pull and push

apb login                              # store your user API key (apb_...)
apb push --apply                       # skills + MCP servers + manifest → hosted playbook
apb pull <guid> --apply                # teammates pull it into their projects
apb sync --target=claude,codex --apply # …and into whichever tools they use

Skills and MCP server definitions travel in both directions. Pull drops them into the portable store (.agents/skills/, .agents/mcp.json); the follow-up sync fans them out to every platform your teammate uses — even if that's a different editor than yours. That's the point: the playbook is the portable unit, not the tool.

Two details we got asked about immediately. First, the hosted side knows things a local file cannot express — request timeouts, auth configuration, curated tool lists. A push updates the connection and leaves all of that alone; it never flattens the richer record into the poorer one. Second, on a brand-new machine the portable store is the only thing on disk and it isn't a deployment target, so sync tells you which agent tools it found for your user and what to pass to --target. No silent no-ops.

Secrets: the contract travels, the credential doesn't

This is the part people expect to be hand-wavy, so to be explicit: no secret value ever moves. What moves is the requirement. sync collects every environment reference in your MCP configuration into the manifest:

"secrets": [
  { "name": "DEPLOY_API_KEY", "ref": "env:DEPLOY_API_KEY", "required": true }
]

A teammate who pulls the playbook now knows exactly which variables to set, and nobody sent a key over anything. Point an entry at a vault instead and your edit survives the next sync. Literal credentials get flagged by doctor, and push refuses to run until they're replaced by references — including credentials sitting in an MCP header or URL.

Install it as a Claude Code plugin

The CLI package is itself a Claude Code / Claude Cowork plugin, with an agentplaybooks skill and slash commands:

/plugin marketplace add integrityauthority/agentplaybooks
/plugin install agentplaybooks@agentplaybooks

Then just ask: "audit my agent config", "make my Claude skills available in ChatGPT and Cursor", or run /agentplaybooks:doctor. The skill knows the safe workflow — plan, show you the diff, apply only after you approve.

Get started

git clone https://github.com/integrityauthority/agentplaybooks
node agentplaybooks/packages/cli/bin/agentplaybooks.js doctor .

Read the full guide in the CLI & Editor Plugins docs, and tell us which platform adapter you want next — ROS 2 is already on the roadmap.