Table Skills: project-handoff
Table Skills (project-handoff)
The project-handoff skill saves goals, decisions and the next step to PROJECT_STATE.md for a fresh Codex or Claude Code session
- 453 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill writes `PROJECT_STATE.md` into the project and rereads it on restore, so dubious entries in it are worth checking
- The optional hooks run Python scripts on each message and on context compaction
- In Codex, with your confirmation, the skill creates a new task and sends it messages
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/table-skills-project-handoffDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill writes `PROJECT_STATE.md` into the project and rereads it on restore, so dubious entries in it are worth checking
- The optional hooks run Python scripts on each message and on context compaction
- In Codex, with your confirmation, the skill creates a new task and sends it messages
This is third-party code. Review the repository files before installing.
What it does
The repository has two sections, and only one is active: project-handoff. For a long task, the skill assembles material to hand over to a new session: the goal, decisions made, key files, real progress and the next step. It records only what affects continuing the work, and the agent does not copy the whole chat. The material goes to `PROJECT_STATE.md` in the project root or into the project's existing state store. If you ask only to save, the skill returns the file path and text for the first message in the new session. With direct permission to create a new task and continue in Codex, the skill creates the session, checks what was recorded and carries on from the first step. The second function: the skill judges whether it is time to suggest a handover. It looks for a safe moment, a clear next step and a concrete benefit from a fresh session. The suggestion comes at the start of the final answer. Optional hooks count automatic context compactions and remind after the third. The neighboring web-stand-in section is declared deprecated by the author and kept as notes on the idea; it is not meant to be installed.
Who it is for. Developers and researchers who run long tasks in Codex or Claude Code, often hit context compaction and lose decisions when switching sessions.
Good fit when
- A session has grown, and before moving to a new one you need to record what was decided and what was rejected
- Work stopped halfway and a fresh-context session will pick it up tomorrow
- The agent has compacted context several times, and you want a prompt for when to switch sessions
- A project has several tasks, and each one's state should sit in its own block of a shared file without overwriting the others
Not a fit when
- You want persistent memory across all projects: the skill saves the state of one task, and AI Memory Skillpack covers shared memory
- You want to store the full chat history: the skill keeps only what affects continuing the work
- You want automatic continuation in Claude Code: it cannot yet create and open a new task, so only saving and the first-message text are available
- You want to carry over installs, commits or publishing: saving has nothing to do with them and does not perform them
Example request
Save the task's progress with project-handoff and give me the first-message text for a new session; do not create the session yetLimitations
The author targets Codex on Windows, where the hook was tested. The Claude Code adapter (`scripts/claude_hook.py`) has shown compaction counting and state injection at work, while the full reminder chain was run on simulated events. It never reached the third real auto-compaction in a live Claude Code session.
The hook is enabled separately: you replace the template paths, merge the configuration with your own and confirm trust in the client. The skill does not add the hook itself and cannot supply missing tools or permissions.
The instructions are in Chinese, and the agent explains them as it goes. Copy the whole skill folder: `SKILL.md` alone is not enough.
How to disable. Delete the project-handoff folder from the agent's skills directory and remove the added hook entries from the Codex config or `~/.claude/settings.json`. To silence reminders for the current task, tell the agent "本任务不再主动提醒交接". The `PROJECT_STATE.md` file stays and is deleted by hand.
FAQ
How do I call project-handoff in Codex and Claude Code?
In Codex with `$project-handoff`, in Claude Code with `/project-handoff`. A plain phrase also works: "save the progress" or "open a new session and continue". A vague request to hand the task over leads only to saving.
Does the skill create a new session without permission?
No. Reminding, saving and creating a new session are separate actions, and the last needs your confirmation of a concrete plan: save, open a new session in the named folder, check and continue. A plain "continue" does not count as permission.
Where does the skill write the state?
To `PROJECT_STATE.md` in the project root or to the project's existing state store. Each task's block is wrapped in BEGIN and END marks with a handover number so several tasks do not overwrite each other.
When does the skill remind about a handover?
After the third automatic context compaction, if the moment is safe and the next step is clear. Later reminders need a new stage, a concrete benefit and three more compactions since the previous one. Without the hook, reminders depend on the agent's self-check.
Should I install web-stand-in?
No. The author declared that section deprecated: only the idea file `web-stand-in/思路.md` remains, with no installation entry.
Discussion
No questions or reviews yet.
Related
SkillsA skill that keeps project memory in plain Markdown files: session logs, a wiki, rules, a to-do list and a note on where to resume
A skill that cross-examines a plan before work starts: numbered questions in rounds with a recommended answer and no changes until you confirm
cafe3310's personal skill collection: spec-driven project work, reviews, reports, browser, 3D and video in one repository
Seven Claude Code skills: turn a video, article or PDF link into text and an action plan, plus a Scrum coach, a session log and an unblock helper
The CLI turns a coding-agent session into a shareable page with conversation, tool activity, files and a diff.
About a hundred skills for Claude Code: meetings, research, publishing, TDD, Telegram, email and personal analytics
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Table Skills: project-handoff (this card) | no | 453 | |
| AI Memory Skillpack | no | 491 | keeps persistent project memory in docs/ai; project-handoff saves one task's state for the next session |
| Claude Code Memory Setup: Obsidian-based memory | no | 1,059 | memory across sessions through an Obsidian vault and /save and /resume commands; here a state file and separate permissions |
README in short
The page, in Chinese, describes the repository as a collection of working AI techniques. One section is active, project-handoff, while web-stand-in is kept as notes. Installation by link for an agent, sample phrases for saving, handing over and restoring, reminder rules, a description of optional hooks, and a request to report the host, system and steps with personal data removed when something breaks follow.
README badge
Original SKILL.md
--- name: project-handoff description: 项目交接：把长任务的目标、有效决定、真实进度、关键文件和下一步整理成接手材料，让新对话核对后继续，避免重述和两边同时改。用户说“保存一下进度”“整理交接”“开个新对话接着做”“读 PROJECT_STATE 继续上次”时使用；宿主注入 project-handoff 压缩计数、到达压缩检查点、准备交付一个可验收阶段、发现已核实的目标或版本混淆时，也用它评估是否建议交接。提醒、保存、新建接续是三个独立动作，互不授权。 --- # 项目交接 让接手对话拿到当前成果、有效决定和明确的第一步，不用用户重述，也不会两个对话同时改。适用于代码、文档、研究和媒体制作。提醒、保存、新建接续是三个动作，各有各的授权：提醒不写文件，保存不开新对话，新建只在用户确认了具体提议之后。