OthmanAdi / planning-with-files

Crash-proof file-based planning that lets AI coding agents survive context loss and complete long-running tasks deterministically across 60+ agent platforms.

活跃维护 MIT Shell Tracked
26.4k 2.2k 9 小时前
CI
claude claude-code claude-skills manus agent-skills planning autonomous-agents codex cursor multi-agent-systems coding-agent context-engineering github-copilot llm-agents long-running-agents claude-code-skills context-rot session-recovery hermes-plugin hermes-skill

星标趋势

数据积累中,暂无足够数据生成趋势图

AI 分析

项目摘要

planning-with-files is a persistent file-based planning skill for AI coding agents that keeps task plans, findings, and progress on disk, enabling agents to survive context loss, /clear commands, and crashes. It features an opt-in completion gate that holds agents until plans are actually completed, and works across 60+ AI coding agents via the SKILL.md standard.

为什么值得关注

With 24.5K stars and rapid adoption, this project solves a critical pain point in AI agent workflows — context loss during long-running tasks. Its cross-agent compatibility via SKILL.md, deterministic completion gate, and Manus-inspired approach make it a practical, battle-tested solution for production agentic coding workflows.

优势

  • Cross-agent compatibility with 60+ AI coding agents via SKILL.md standard
  • Crash-proof persistent state that survives context loss and /clear commands
  • Opt-in completion gate for deterministic task completion verification
  • Strong validation with 96.7% benchmark pass rate, A/B blind testing, and security audit
  • Very active development with 72 releases in 6 months and 46 contributors

局限性

  • File-based approach may have scalability limits for highly complex distributed multi-agent scenarios
  • Dependent on SKILL.md standard adoption across agent ecosystems
  • Primarily focused on coding tasks, may require adaptation for non-coding agent workflows

使用场景

  • Long-running autonomous coding sessions that must survive context window limits
  • Multi-agent coordination with shared persistent state on disk
  • Crash-proof agent workflows that resume after interruptions
  • Deterministic task completion verification with the completion gate pattern
目标用户: Developers and teams using AI coding agents (Claude Code, Cursor, Copilot, Codex CLI, etc.) who need reliable long-running agent workflows with persistent state and deterministic completion.
学习曲线:
分析模型:LongCat-2.0 | 分析时间:1 个月前