claude-runtime

What problem does it solve?

This Skill reduces failed or inconsistent Claude invocations by standardizing how Claude Code subprocess output is parsed and how Claude SDK/MCP imports are performed across local and Railway environments.

Core Features & Use Cases

• Subprocess-first Claude Code streaming: Ensures the system parses stream-json events like system, assistant (including tool_use), tool_result, and result for deterministic runtime behavior.
• Environment-aware execution paths: Directs teams to use the claude -p subprocess only when the Claude Code CLI is available, and to switch to ANTHROPIC_API_KEY with claude_agent_sdk when running in cloud where the CLI is not present.
• Correct MCP SDK import paths: Prevents runtime failures by requiring the specific MCP SDK subpath imports for McpServer and stdio transport objects.
• Guardrails for rejected SDK migration: Documents why the Managed Agents API path was not adopted due to PoC failures, helping avoid repeated dead-ends.

Quick Start

Ask your agent code to follow the claude-runtime rules so it selects the correct Claude invocation mode for the current environment and parses tool and result events consistently.