ParaContinuity

English | 中文

Overview

ParaContinuity is a lightweight Codex skill for keeping long-running AI-assisted projects handoff-ready without creating corporate-process bloat.

It is designed for projects where one giant now.md, handoff.md, or chat-export file has become too large for every new session to read safely. The skill teaches an AI agent to split project memory into a small documentation tree:

• current state
• recent activity and discovered local TODOs
• future roadmap
• durable decisions
• executable runbooks
• data/interface contracts
• experiment results
• short handoff notes
• historical archive

The core rule is simple:

Docs preserve intent and constraints. Harnesses, scripts, tests, and reports validate behavior.

When To Use

Use ParaContinuity when:

• A project spans many AI sessions, accounts, machines, or people.
• now.md or a similar handoff file has become too large.
• You need a new AI agent to understand what to read first.
• You want reproducible handoff without simulating a large corporate workflow.
• You need to preserve why a technical choice was made, not just what code exists.
• You want clear update rules for progress, roadmap, decisions, commands, schemas, and experiments.
• You recently discovered a data gap, partial migration, or local TODO that future sessions must not lose.

What Is Included

ParaContinuity/
  SKILL.md
  agents/
    openai.yaml
  README.md
  LICENSE

SKILL.md is the actual Codex skill. README.md is this Hugging Face card.

Recommended Docs Tree

When the skill is applied to a project, it recommends:

docs/
  HANDBOOK.md        # Entry point and operating rules.
  CURRENT.md         # Current state, active constraints, next actions.
  RECENT.md          # Recent activity, discoveries, data gaps, local TODOs.
  ROADMAP.md         # Future direction and milestones.
  DECISIONS.md       # Durable choices and why they were made.
  RUNBOOKS.md        # Commands, environment, operational procedures.
  DATA_CONTRACTS.md  # Schemas, data formats, interface contracts.
  EXPERIMENTS.md     # Results that affect future decisions.
  HANDOFF.md         # Short latest handoff for session/account/person switches.
  archive/           # Historical long notes and stale reports.

Installation

Copy the skill folder into a Codex-discoverable skills directory, or keep it inside a repo and explicitly point the AI agent to it.

Typical local install:

mkdir -p ~/.codex/skills
cp -r ParaContinuity ~/.codex/skills/project-continuity

Or use it from a project repo:

Use the project-continuity skill at docs/skills/project-continuity to set up lightweight handoff docs for this repo.

Example Prompt

Use $project-continuity for this repo.

Do not read the whole now.md first. Search for recent/current sections only if needed.
Create or update a lightweight docs tree with:
docs/HANDBOOK.md
docs/CURRENT.md
docs/RECENT.md
docs/ROADMAP.md
docs/DECISIONS.md
docs/RUNBOOKS.md
docs/DATA_CONTRACTS.md
docs/EXPERIMENTS.md
docs/HANDOFF.md
docs/archive/

Keep docs focused on intent, constraints, recent discoveries, durable decisions, executable commands, data contracts, experiment results, and restart instructions.
Do not write secrets into docs.

Example Workflow

1. The agent scans top-level files and existing handoff notes.
2. It avoids reading huge history files end to end.
3. It creates or updates the standard docs/ tree.
4. It writes recent handoff-relevant discoveries into RECENT.md.
5. It writes durable decisions into DECISIONS.md.
6. It writes commands into RUNBOOKS.md.
7. It writes schemas into DATA_CONTRACTS.md.
8. It writes the shortest restart note into HANDOFF.md.

RECENT.md Rule

RECENT.md is for facts that are easy to lose between sessions but are not yet durable decisions:

• important actions completed in the last 7-14 days
• discovered data gaps
• partial migrations
• local TODOs that affect the next handoff
• artifact paths that future agents must inspect soon

Promote a fact from RECENT.md into CURRENT.md, HANDOFF.md, DECISIONS.md, or RUNBOOKS.md when it becomes current state, a first-screen handoff item, a durable decision, or a repeatable command.

Design Principles

• Keep current-state docs short.
• Archive history instead of making every future agent read it.
• Use scripts and reports as validation, not prose confidence.
• Record non-obvious intent because code often cannot recover it.
• Avoid adding process for its own sake.
• Never store API keys or secrets in project docs.

---

中文

概述

ParaContinuity 是一个轻量级 Codex skill，用来让长期 AI 协作项目保持可交接，而不是把文档流程做成复杂的公司制度。

它适用于这种情况：项目里只有一个巨大的 now.md、handoff.md 或聊天记录文件，新 session 每次读完都快把上下文撑爆，而且进度、路线、决策、命令、schema 和实验结果都混在一起。

这个 skill 会让 AI agent 把项目记忆拆成一棵小型文档树：

• 当前状态
• 最近动作、发现和局部待办
• 未来路线
• 长期决策
• 可执行命令
• 数据/接口契约
• 实验结果
• 短交接说明
• 历史归档

核心原则是：

文档保存意图和约束；harness、脚本、测试和报告负责验证。

什么时候使用

当你的项目出现这些情况时，适合使用 ParaContinuity：

• 项目跨很多 AI session、账号、机器或真人交接。
• now.md 或类似交接文件已经太长。
• 你需要新 AI 知道先读什么、不要先读什么。
• 你想要可复现的交接，但不想模拟复杂公司流程。
• 你需要保存“为什么这么做”，而不只是保存代码。
• 你想给进度、路线、决策、命令、schema、实验结果制定清晰更新规则。
• 你刚发现了数据缺口、半迁移状态或局部待办，不希望下一次 session 忘掉。

包含内容

ParaContinuity/
  SKILL.md
  agents/
    openai.yaml
  README.md
  LICENSE

SKILL.md 是真正给 Codex 使用的 skill。README.md 是 Hugging Face 页面展示的 card。

推荐文档树

应用到项目时，推荐生成：

docs/
  HANDBOOK.md        # 入口和操作规则
  CURRENT.md         # 当前状态、约束、下一步
  RECENT.md          # 最近动作、发现、数据缺口、局部待办
  ROADMAP.md         # 未来路线和里程碑
  DECISIONS.md       # 长期有效的技术取舍和原因
  RUNBOOKS.md        # 命令、环境、运行和恢复流程
  DATA_CONTRACTS.md  # schema、数据格式、接口契约
  EXPERIMENTS.md     # 会影响后续决策的实验结果
  HANDOFF.md         # session/账号/人员切换用的短交接
  archive/           # 历史长文档和过时报告

安装方式

可以把 skill 文件夹复制到 Codex 能发现的 skills 目录，也可以放在项目仓库里，然后在提示词中明确指向它。

典型本地安装：

mkdir -p ~/.codex/skills
cp -r ParaContinuity ~/.codex/skills/project-continuity

也可以在项目内使用：

请使用 docs/skills/project-continuity 里的 project-continuity skill，为这个 repo 建立轻量级交接文档。

示例提示词

请使用 $project-continuity 接手这个 repo。

不要先完整读取 now.md。只有需要历史细节时，才搜索 now.md 的最近/current 相关段落。
请建立或更新轻量级 docs tree：
docs/HANDBOOK.md
docs/CURRENT.md
docs/RECENT.md
docs/ROADMAP.md
docs/DECISIONS.md
docs/RUNBOOKS.md
docs/DATA_CONTRACTS.md
docs/EXPERIMENTS.md
docs/HANDOFF.md
docs/archive/

文档只记录意图、约束、近期发现、长期决策、可执行命令、数据契约、实验结论和接续说明。
不要把 API key 或 secret 写进文档。

示例流程

1. AI agent 扫描项目顶层文件和已有交接文档。
2. 避免从头到尾读取巨大历史文件。
3. 创建或更新标准 docs/ 文档树。
4. 把最近发现、数据缺口、局部待办写进 RECENT.md。
5. 把长期技术取舍写进 DECISIONS.md。
6. 把可重复命令写进 RUNBOOKS.md。
7. 把 schema 和接口约束写进 DATA_CONTRACTS.md。
8. 把最短接续说明写进 HANDOFF.md。

RECENT.md 规则

RECENT.md 用来记录容易在 session 切换时丢失、但还不一定是长期决策的事实：

• 最近 7-14 天完成的重要动作
• 新发现的数据缺口
• 半迁移/半完成状态
• 会影响下一次接手的局部待办
• 近期必须检查的产物路径

当某条记录变成当前真实状态、接手第一屏事项、长期技术取舍或可重复命令时，把它同步提升到 CURRENT.md、HANDOFF.md、DECISIONS.md 或 RUNBOOKS.md。

设计原则

• 当前状态文档要短。
• 历史内容归档，不让每个新 agent 都读。
• 用脚本和报告验证，不靠文字保证。
• 记录非显然意图，因为代码通常不能稳定反推出业务取舍。
• 不为了流程而增加流程。
• 永远不要把 API key 或 secret 写进项目文档。