Claude Managed Agents Skill

This skill gives Claude Code deep, always-available knowledge of Anthropic's Managed Agents API — the $0.08/hour infrastructure for running Claude as an autonomous agent in cloud containers. Built by 0xArx against the managed-agents-2026-04-01 beta header. The skill focuses on patterns and gotchas that aren't obvious from reading API documentation. Stream ordering (open the event stream before sending the message). Stop reason handling on idle status. MCP authentication architecture that separates server URLs from vault-stored credentials. Agent version locking requirements. These are the things that cost developers hours of debugging. Architecture uses progressive disclosure across three tiers. The skill name and description sit in context at zero cost when idle. SKILL.md loads in full when the skill activates — covering core concepts, the build flow, and event loop patterns. Seven reference files load on-demand for deep dives: events, tools, vaults and MCP, container specs, session management, research preview features, and production patterns. Installation is a one-liner: curl the repo into ~/.claude/skills/. Claude Code discovers it automatically — no restart required. The skill activates when you reference managed agent endpoints, build event loops, configure MCP connectors, work with tool permissions, or design long-running autonomous tasks. It covers mid-session GitHub token rotation, custom tool event handling to prevent silent hangs, and session deletion's effect on event stream termination. Requires Claude Code and an ANTHROPIC_API_KEY with managed agents beta access. Research preview features (outcomes, multiagent, memory) need the separate managed-agents-2026-04-01-research-preview header.