Skills Library
modular procedures, any runtime
Save the procedures your agents repeat. Attach them to tasks or agents. Export to Claude, Cursor, Codex on demand. Trigger-based routing means the right skill loads when it's relevant.
--- name: react-component-builder description: Use when creating new React components. Generates them with TypeScript, Tailwind, and colocated tests. allowed-tools: Read, Edit, Bash --- # React component builder When asked to create a new component: 1. Place it in `src/components/<Name>.tsx` unless told otherwise. 2. Use `forwardRef` if it accepts a `ref` prop. 3. Type props with an exported interface named `<Name>Props`. 4. Style with Tailwind utility classes; no `@apply` outside of `globals.css`. 5. Add a colocated test `<Name>.test.tsx` covering the happy path and one edge case. Avoid: - `React.FC` (prefer typed function declarations) - Index re-exports unless explicitly requested - Inline objects in JSX props (extract to `useMemo` if needed)
One library, every runtime
The Skills Library is built on Claude's SKILL.md spec, but compiles to whatever your team uses today.
Trigger-driven
Each skill has a 'use when' description. Claude reads the descriptions of every skill in scope and routes to the right one automatically, no manual selection.
Attach to tasks or agents
Pre-attach a skill set to an agent so its toolkit follows every task it runs. Or pin specific skills to a single backlog ticket. Both surfaces have a sticky default.
Universal export
Save once, push to Claude (.claude/skills/), Cursor (.cursor/rules/), Codex (AGENTS.md), Windsurf or Aider in one click. Switch runtimes without rewriting your skills.
Multi-file skills
A skill is a folder, not a single file: ship helper scripts, reference documents and templates next to SKILL.md. Import an existing skill folder as it is, and the files are unpacked on disk when an agent starts, so it can read and run them.
Version history
Every save that changes a skill keeps the previous version, bundled files included. Compare an earlier version with the current one and restore it in one click, from the desktop or the mobile app.
Update from git
Link a skill to a git repository, with an optional branch or tag and a subfolder when one repository holds several skills. Update pulls the latest SKILL.md and its files using your own git access, and the previous version stays in the history.
Sync across devices
Project skills follow your repo. Account-level skills follow you everywhere. Desktop and mobile both read, write and publish: the phone is not a read-only viewer.
Allowed-tools whitelisting
Constrain a skill to a subset of tools (Read, Edit, Bash…). The runtime enforces the whitelist when the skill is loaded: sharper boundaries, fewer surprises.
Personal skills, gitignored
Mark a skill as personal and it stays on your machine. Project-level skills go into git so the team gets them; your private ones don't leak.
How it works
Open the Skills Library
Click the Skills tile in the project sidebar (or hit it from the agent terminal on mobile). Browse project + global tabs, search, organize in folders.
Write the trigger first
The 'description' field is what Claude reads to decide whether to load the skill. Phrase it as 'Use when…' or 'Use to…'. Two clear sentences beats a long paragraph.
Attach it to a task or agent
On task creation, multi-select skills next to the agent picker. Or set defaults per agent so its toolkit travels with it. Last set is sticky for the next task.
Skills are injected at task launch
When the task transitions to in_progress, the body of each skill is appended to the agent's first message in native SKILL.md format. Claude treats it like a disk-loaded skill.
Optional: export to disk
Click 'Save to >' on any skill to write a real file: Claude under .claude/skills/<name>/, Cursor under .cursor/rules/, Codex appended to AGENTS.md, etc.
.claude/skills/<name>/SKILL.mdWrite once, deploy to any agent
Skills are stored in AgentsRoom in their canonical (Claude SKILL.md) form. When you click "Save to", AgentsRoom compiles them into the format your target runtime expects, automatically.
Claude Code
.claude/skills/<name>/SKILL.mdCursor
.cursor/rules/<name>.mdcWindsurf
.windsurf/rules/<name>.mdOpenAI Codex
AGENTS.md (managed block)Aider
CONVENTIONS.md (managed block)Generic
<name>.mdFAQ
What's the difference between a skill and a prompt?
A prompt is text you copy into the agent. A skill is a procedure with a 'use when' description that the runtime loads ON DEMAND when relevant. Skills follow you across tasks; prompts are imperative one-shot inputs.
Does this only work with Claude Code?
Storage is Claude-format (SKILL.md), but the export menu compiles to Cursor, Windsurf, Codex, Aider and plain markdown. Any agent that consumes one of these runtimes can use your skills.
How are skills delivered to the agent?
When a task moves to in_progress, the bodies of attached skills are injected into the agent's initial prompt as native SKILL.md blocks. The agent reads them the same way as on-disk skills.
Can I share skills with my team?
Project skills sync via the AgentsRoom API to every teammate on the same project. Skills marked Personal stay on your machine.
Can I import skills I already wrote?
Yes. The Import button scans .claude/skills/, .cursor/rules/, and .windsurf/rules/ in your project (or user home for Claude), parses the YAML frontmatter, and adds matches to your AgentsRoom library.
Can a skill apply to a whole project instead of one agent?
Yes. Skills stack in three layers: a project baseline that every agent of that project inherits, the defaults attached to a given agent, and the selection pinned on a single backlog ticket. The baseline is the one to use for conventions that must never be forgotten, like your commit rules or your review checklist, since it applies whatever agent picks up the work.
Where do I find skills written by other developers?
In the Community tab of the Skills Library, without leaving AgentsRoom. Search the public catalog, filter by category, read a skill in full, then either use it as it is or copy it into your library, choosing whether it lands in this project or in your account. Publishing yours is a checkbox in the skill editor, unticked by default, and every submission is moderated before it appears.
Can I use my skills from the mobile app?
Yes. The library is synced to the mobile companion through the end-to-end encrypted relay, so you can attach a skill to a task from your phone the same way you would on the desktop.
Can I keep a skill away from every agent unless I attach it?
Yes. Mark a skill "Attached agents only" in the editor and it disappears from the library agents browse over MCP: only the agents you attach it to (as an agent default, in the project baseline or on a ticket) can see and load it. A big library stays usable without every specialised procedure being picked up whenever a request sounds similar.
Can I keep a skill in sync with a git repository?
Yes. In the skill editor, open Git source and enter the repository URL, an optional branch or tag, and the subfolder that contains SKILL.md if the repository holds several skills. Fetch or Update reads SKILL.md and its sibling files with the git access of your computer (SSH key or credential helper), shows the result in the editor, and saving keeps the previous version in the history so an update can always be undone.
You may also like
Prompt Library
Save the prompts you reuse, organize them in folders, and send them to any agent in one click. Same DNA as the Skills Library.
Teams (multi-agent workflow)
Chain multiple agents into a pipeline. Skills attached to each agent travel with the handoff.
AgentsRoom MCP
The MCP server agents use to read/save skills, prompts, and dev commands at runtime, no extra plumbing.
Backlog & Task Board
Attach skills directly to backlog tickets. The agent loads them when it picks up the task.
Further reading
Claude Ads: the Claude Code skill that audits your ad accounts
Claude Ads is an open source skill for Claude Code: 250+ checks on Google, Meta, LinkedIn, TikTok or Amazon Ads, a score out of 100 and a prioritized action plan, in about ten minutes. Install, commands, limits, and how to orchestrate it in AgentsRoom.
AGENTS.md: One Context File for Every Coding Agent (Codex, Antigravity, Claude)
AGENTS.md is the portable instructions file your AI coding agents read before touching your code. What to put in it, how it differs from CLAUDE.md, and how to keep one context across Codex, Antigravity and Claude.
Build your skill set
Download AgentsRoom and start your Skills Library in 30 seconds. Free for up to 2 projects.
Companion app: monitor your agents on the go
Bring your own: Claude, Codex, Antigravity CLI, or other AI provider.
Push bugs and requests straight to your public backlog.