Harness engineering for AI session continuity — cross-tool handoff governance for Codex, Claude Code, and Gemini CLI. PLAN→READ→CHANGE→QC→PERSIST workflow with anti-drift guardrails.

unknownagents-mdai-codingai-handoffclaude-code

推荐理由
README 将它定位为「Harness engineering for AI session continuity — cross-tool handoff governance for Codex, Claude Code, and Gemini CLI」,核心痛点是让 AI agent 通过可复用 skills 和工具流程处理具体任务。它有一定社区验证,同时仍保留发现潜力,适合作为「AI agent 工具链」的候选项目。

注意事项
license 信息不够明确,采用前要确认授权边界;项目还偏早期,需要重点验证核心路径是否稳定;README 摘要信息有限,发布前建议再人工扫一遍文档;展示前建议跑通 README quickstart,并确认部署成本、外部依赖和数据安全边界。

这个推荐有帮助吗?

README 摘要

Harness engineering for AI session continuity — cross-tool handoff governance for Codex, Claude Code, and Gemini CLI. PLAN→READ→CHANGE→QC→PERSIST workflow with anti-drift guardrails.

Stars
27
Forks
5
Last push
2026/5/8

相似项目

microsoft/PromptKit

Agentic prompts are the most important code you're not engineering. PromptKit fixes that — composable, version-controlled prompt components (personas, protocols, formats, templates) that snap together into reliable, repeatable prompts for bug investigation, design docs, code review, security audits, and more. Works with any LLM.

GitHub
52
stars
JavaScriptmitagentic-aiaicode-reviewcomposable

推荐理由:README 将它定位为「Agentic prompts are the most important code you're not engineering」,核心痛点是把 prompt 技巧、模板和工作流沉淀成可复用资产。它有一定社区验证,同时仍保留发现潜力,license 清晰,主要技术栈是 JavaScript,适合作为「AI agent 工具链」的候选项目。

注意事项:项目还偏早期,需要重点验证核心路径是否稳定;README 摘要信息有限,发布前建议再人工扫一遍文档;展示前建议跑通 README quickstart,并确认部署成本、外部依赖和数据安全边界。

这个推荐有帮助吗?
wuyoscar/gpt_image_2_skill

GPT Image 2 prompt gallery, image prompt library, agentic skill, and CLI for OpenAI image generation/editing

GitHub
1.7K
stars
Pythonmitagent-skillsai-image-promptsclaude-code-skillcli

推荐理由:README 将它定位为「GPT Image 2 prompt gallery, image prompt library, agentic skill, and CLI for OpenAI image generation/editing」,核心痛点是把图像生成、编辑或多模态能力接入真实创作工作流。它有一定社区验证,同时仍保留发现潜力,license 清晰,主要技术栈是 Python,适合作为「AI 绘图和多模态工作流」的候选项目。

注意事项:README 摘要信息有限,发布前建议再人工扫一遍文档;展示前建议跑通 README quickstart,并确认部署成本、外部依赖和数据安全边界。

这个推荐有帮助吗?