Agent Handoff Kit keeps progress, key decisions, file locations and unfinished tasks in your project so the next AI can find them.
This guide includes updates that have not been released yet. Features available to download depend on the published version.
Without a handoff, you often have to repeat context, find files again and remind the AI which approaches were already tried. With the tool, important context stays in the project. You still set requirements and make decisions; the AI organizes and checks the work.
Long tasks can lose past decisions, your corrections and unfinished items between conversations. The tool helps the AI save them. Next time, it reads the full handoff before explaining where things stand.
Resume work →New articles, reference PDFs and work plans should not live only in one conversation. The tool records what files are for, when to read them and what related material needs updating, reducing missed sources and wrong-version use.
Check file connections →Alongside current progress, the tool keeps important decisions, their reasons and work records. Ask “Why did we choose this?” and the AI can trace the saved evidence. Long records are organized and archived as needed, without rereading everything every time.
Explore history →Open the same folder in a supported AI tool to reuse saved handoffs and working requirements. Project content is handled separately from official tool files; upgrades preserve custom changes and stop at conflicts.
Keep a working requirement →You do not select work modes or memorize rules. Describe the result you want; the AI applies the relevant approach and asks only when missing information affects the outcome.
Check the audience, purpose and tone, then produce content that fits the use.
Check sources and dates, separating facts, inferences and remaining uncertainty.
Read relevant files, preserve existing changes and check the actual result afterward.
Distinguish sources, notes and deliverables; verify connections and permissions, then read back any writes.
Enter shortcuts in your project’s AI conversation, not in the terminal. Claude Code, Gemini CLI and Antigravity CLI use /; Codex uses $, or selection through /skills.
Installation and upgrades include them for the current project only, with no separate activation. Select a shortcut for examples and details.
See all commands: guidance, files, working requirements, checks and updates →
Use handoff-kit-progress to see where the project stands, what remains and how it got here. Overview shows current work; History lets you find changes by date; Documents shows the purpose and connections of important files.
Actual interface screenshot; Atlas is a fictional example. Select the image to enlarge it. Each project shows its own records.
The page updates automatically when the AI saves records. Asking the AI to open it and saving records use some tokens; viewing, filtering and subsequent page updates do not make further AI calls. Unsaved work does not appear.
Dashboard help →Open your project folder in your AI tool and paste the sentence below. The AI confirms the folder, installs or upgrades, then checks the result. You do not choose the installation commands.
Read https://adamchanadam.github.io/agent-handoff-kit/agent-handoff-kit-ai-install.en.html and install or upgrade Agent Handoff Kit in this project folder.
You need an AI tool with read and write access to your project folder, plus Node.js 18 or later. Web chat without folder access is not supported; ask the AI to check if unsure. Use the same sentence for older versions or existing AI work files.
The AI must read the full handoff before claiming it has resumed. Start on its own reports the current state; work begins when you give a task. Wrapping up can save unfinished work without marking the whole project complete.
Before uploading, publishing, deleting important data or changing access, the AI must explain the effects and obtain the required approval. Passwords and secret keys must not enter handoff records. The tool’s rules and checks help; they cannot guarantee an AI will never make a mistake.
After using external tools, the AI should close only temporary processes it can confirm belong to this task. If it is unclear which task a process belongs to, it asks first to avoid interrupting other work. Missing permissions or locked files must be reported, not bypassed.
The tool carries project work between conversations. To set the AI’s reply style and general working habits separately, you can also use Adam-AI-Instructions. It is optional.