claude-code-workflow  by runesleo

AI development workflow for persistent memory and context

Created 1 month ago
539 stars

Top 58.8% on SourcePulse

GitHubView on GitHub
Project Summary

A battle-tested workflow template for Claude Code, designed to overcome its inherent limitations in memory and context management. It transforms Claude Code into a persistent, self-improving development partner by implementing robust memory management, context engineering, and intelligent task routing. The template enables Claude to learn from past mistakes, maintain context across long sessions, and efficiently utilize different model tiers, making it suitable for production-level AI-assisted development.

How It Works

The core architecture employs a three-layer system to optimize context window usage. Layer 0 contains always-loaded rules (behaviors, memory flush triggers). Layer 1 loads on-demand documentation (agents, safety, task routing) as needed, conserving tokens. Layer 2 maintains hot, working memory (daily logs, projects, tasks) for instant recall. This layered approach ensures efficient token management and a persistent, context-aware development session.

Quick Start & Requirements

  • Installation: Clone the repository and copy/symlink its contents into your Claude Code configuration directory (~/.claude/).
  • Customization: Edit ~/.claude/CLAUDE.md to configure user info, project memory routes, and the Single Source of Truth (SSOT) table.
  • Requirements: Claude Code CLI (Claude Max or API subscription). Optional: Codex CLI, Ollama for local models.
  • Links: GitHub Repo: https://github.com/runesleo/claude-code-workflow. Telegram Community mentioned.

Highlighted Details

  • SSOT (Single Source of Truth): Enforces canonical data locations, preventing inconsistencies.
  • Memory Flush: Automatically saves progress on task completion, commits, and exits, ensuring no work is lost.
  • Verification Before Completion: Mandates running verification commands and checking output before task completion, mitigating AI coding failures.
  • Three-Tier Task Routing: Intelligently routes tasks to Opus, Sonnet, Haiku, Codex, or Local models based on complexity and cost.
  • Sunday Rule: System optimization is scheduled for Sundays; other days prioritize shipping code.

Maintenance & Community

Developed by @runes.leo. A Telegram Community is linked. No specific details on sponsorships, partnerships, or deprecations are provided.

Licensing & Compatibility

  • License: MIT License.
  • Compatibility: Permissive MIT license allows commercial use, modification, and distribution, including in closed-source projects, with attribution.

Limitations & Caveats

The README positions this as a production workflow, implying a need for existing Claude Code familiarity. The "Sunday Rule" is a philosophical constraint that can be reconfigured but is presented as a core principle. No explicit technical limitations or known bugs are detailed.

Health Check
Last Commit

1 month ago

Responsiveness

Inactive

Pull Requests (30d)
1
Issues (30d)
1
Star History
265 stars in the last 30 days

Explore Similar Projects

Starred by Christian Laforte Christian Laforte(Distinguished Engineer at NVIDIA; Former CTO at Stability AI), Dan Guido Dan Guido(Cofounder of Trail of Bits), and
4 more.

ruflo by ruvnet

4.8%
31k
AI orchestration platform for complex development workflows
Created 10 months ago
Updated 3 days ago
Feedback? Help us improve.