Skills real projects publish on GitHub, most starred first. Each one says what it will make an agent do before you copy it.
AnthropicSkillanthropics/claude-quickstarts/managed-agents/assistant-ui/skill.md
Things that aren't obvious from the docs and tend to cost debugging time. If you're Claude driving setup for a user, walk the Setup checklist and use the rest for reference when something looks off.
AnthropicSkillanthropics/claude-quickstarts/managed-agents/chat-sdk/skill.md
Things that aren't obvious from the docs and tend to cost debugging time. Every browser conversation is a DM on a durable thread (web:{userId}:{conversationId}), and in this quickstart the conversationId half is a Managed Agents session ID, verbatim. The page creates the session first (POST /api/sessions → sessions.create()), hands the returned ID to useChat as its threadId, and every message in that conversation lands in that session. There is no mapping table: the bridge decodes the session ID back out…
AnthropicSkillanthropics/claude-quickstarts/managed-agents/copilot-kit-ag-ui/.claude/skills/verify/SKILL.md
Build and drive the copilot-kit-ag-ui demo to verify changes end-to-end.
AnthropicSkillanthropics/claude-quickstarts/managed-agents/knowledge-wiki/skill.md
Follow these steps in order when the user asks you to set up or run this cookbook. Confirm each step's output before moving to the next. Two features must be enabled on the user's organization: Ask the user to confirm both, and point them at <https://claude.com/form/claude-managed-agents> if not. Without dreaming, POST /v1/dreams returns 404 and the notebook fails at the consolidation step. Everything before that still runs, so it's fine to proceed for a partial walkthrough — just say so…
AnthropicSkillanthropics/claude-quickstarts/managed-agents/linear/skill.md
Things that aren't obvious from the docs and tend to cost debugging time. It's an HTTP POST a service sends to a URL you gave it, with a small JSON body describing an event. You register the URL once, and the service calls it whenever the event fires. No polling, no held-open connections. Linear's Agent Platform and Managed Agents don't share a wire format or credentials. Something has to translate "Linear @mention" → "user.message" on the way in, and "session…
AnthropicSkillanthropics/claude-quickstarts/managed-agents/mcp-server-typescript/skill.md
Things that aren't obvious from the docs and tend to cost debugging time. The Claude you're typing to in Desktop is a relay. It calls sendmessage with your words, calls waitfor_idle to block until the managed agent finishes, then shows you the reply. The work itself (tool use, code execution, repo edits) happens in the Managed Agents session, not in Desktop. listagents, getagent, createsession, sendmessage, interrupt, getsession, listevents, archivesession are straight endpoint wrappers. waitfor_idle is the only editorial: MCP is…
AnthropicSkillanthropics/claude-quickstarts/managed-agents/roadtrip-planner/skill.md
Work top to bottom. Every step has a verification before the next one starts. Verify the vendor keys from your own machine before involving the sandbox: Both return JSON. A 403 here is a key problem, not a quickstart problem. ant apply (which setup runs with --yes; needs ant 1.34 or later and jq) creates the environment (networking limited to developer.nps.gov and api.windy.com), the vault, the reviewer agent (Opus, review-only prompt), and the planner agent (bash on, websearch/webfetch off, a…
AnthropicSkillanthropics/claude-quickstarts/managed-agents/sentry/skill.md
Things that aren't obvious from the docs and tend to cost debugging time. An environment_variable vault credential is deliberately the only way to set env vars in a managed sandbox: What this doesn't give you: the agent can still use the credential for anything the allowlisted host permits. A token with event:write lets the agent resolve and modify issues. Pick scopes accordingly. The vault limits where the token can go, and Sentry scopes limit what it can do once there.
AnthropicSkillanthropics/claude-quickstarts/managed-agents/slack/skill.md
Things that aren't obvious from the docs and tend to cost debugging time. Neither carries the agent's output. Both are signals with IDs. Anthropic's session.status_idled payload is deliberately thin: {type, id}. You follow up with sessions.retrieve(id) (metadata) and sessions.events.list(id) (output). Push the signal, pull the data. On kickoff, set metadata: {slackchannel, slackthreadts, slackteam}. When the idle webhook arrives later with only a session ID, retrieve the session, read those keys back, and chat.postMessage to exactly the right thread, with nothing…
A folder with a SKILL.md file: a name, a description of when to use it, and instructions. Claude loads a skill only when the task matches its description.
Copy the folder into your project's .claude/skills/ directory, or into your own skills folder to use it everywhere.
We read each file for commands that read secrets, delete things or pipe downloads into a shell, and say so before you copy it. No warning is not a promise that a file is safe.
Open a skill to see its discussion. Reports from people and their agents are coming.
