Does this skill work with any Claude Code framework?

Yes, it is designed with high portability and uses framework-agnostic principles for state saving, signaling, and resumption.

Will I lose my work progress if a subagent pauses?

No, the protocol mandates saving a comprehensive state or checkpoint before pausing, ensuring the agent resumes exactly where it stopped without re-analyzing code.

Why can't subagents call AskUserQuestion directly?

Most agent frameworks restrict interactive tools to the main conversation thread to prevent multi-threaded UI conflicts; this protocol provides a safe bridge for delegated agents.

How does the agent handle the user's response?

The main conversation acts as an intermediary, capturing the user's answer and passing it back to the subagent as a structured prompt upon resumption.

Can I use this with MCP servers?

Absolutely. The skill includes specific examples for integrating with MCP servers like Memento to handle entity-based state preservation during the pause-resume cycle.

User Input Protocol for Claude Subagents

Name: User Input Protocol for Claude Subagents
Author: jwilger

byjwilger

•

生产力与工作流

Implements a robust pause-and-resume pattern that allows background agents to request user clarification without losing progress or context.

The User Input Protocol solves a critical limitation in agentic frameworks where subagents or delegated background tasks cannot directly trigger interactive user prompts. By providing a standardized architecture for state preservation and structured question formatting, this skill enables long-running tasks to handle ambiguity gracefully. It ensures that when an agent hits a decision point, it saves its current progress into a checkpoint, signals the main conversation to gather user feedback, and resumes seamlessly once input is received, preventing wasted computation and frustrating manual restarts.

主要功能

01Framework-agnostic state preservation and checkpointing patterns

02Comprehensive constraints to prevent background agent tool failures

03Structured JSON question formatting with context and multi-choice options

047 GitHub stars

05Standardized subagent pause-and-resume workflow logic

06Seamless task resumption mechanisms using saved metadata

使用场景

01Background mutation testing agents requiring human decisions on fix task creation

02Domain modeling agents resolving architectural conflicts or ambiguous business rules

03Automated code review agents needing confirmation on complex refactor directions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jwilger/claude-code-plugins user-input-protocol

For use in Claude.ai and ChatGPT

Download Skill