Implements a persistent, file-based workflow for managing complex tasks and long-running autonomous sessions.
Inspired by the workflow patterns of highly successful AI agents, this skill provides a structured framework for Claude Code to manage state on disk rather than relying solely on volatile context memory. It introduces a mandatory documentation system using three core markdown files—task_plan.md, findings.md, and progress.md—ensuring that research, decisions, and error logs are preserved throughout multi-step operations. This approach minimizes information loss during long sessions, enforces a disciplined 3-strike error protocol, and optimizes performance for complex research, software building, or deep troubleshooting tasks.
Key Features
01Mandatory 2-Action Rule to prevent visual and multimodal data loss
02Persistent 'Disk Memory' using markdown files to overcome context limits
03State management for tasks spanning dozens of tool calls
04Standardized templates for task plans, research findings, and session logs
053-Strike Error Protocol for systematic and robust troubleshooting
0611,059 GitHub stars
Use Cases
01Managing multi-step software development projects from scratch
02Handling complex bug fixes that require extensive investigation and trial
03Conducting deep technical research and competitive analysis
