handoff
Health Gecti
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Community trust — 12 GitHub stars
Code Gecti
- Code scan — Scanned 2 files during light audit, no dangerous patterns found
Permissions Gecti
- Permissions — No dangerous permissions requested
Bu listing icin henuz AI raporu yok.
Claude Code skill that quizzes you on the code Claude wrote — verify you understood it before shipping.
handoff is a Claude Code skill
that turns passive review of AI-written code into an active comprehension check.
When a feature is complete, it interviews you with a short quiz — where the
logic lives and why the decisions were made — then reports a score and the
exact spots to re-read.
The name captures the idea: the handoff of the code from Claude to you. You
shouldn't merge what you can't explain.
Table of Contents
- Why
- Features
- Demo
- Installation
- Usage
- Auto-suggestion hook
- Configuration
- Requirements
- Project layout
- Packaging
- Contributing
- License
Why
AI writes code fast. The risk is shipping it without reading it. Tests catch
broken code; nothing catches code you don't understand. handoff closes
that gap by checking that the human who clicks "merge" can actually explain the
change — its location, its rationale, and its behavior.
Features
- 🧬 Hybrid question source — combines conversation context (the why:
decisions, trade-offs, what could break) withgit diff(the where: realfile:linelocations), and degrades gracefully when there's no git. - 📏 Scaled to the change — ~3 questions for a small diff, 5 for medium,
8–10 for a large feature. - 🔀 Mixed question types — multiple-choice for "where / what behavior",
open-ended for "why". - 1️⃣ One question at a time — you answer, it evaluates, then continues.
- 🎯 Wrong-answer flow — points you to the
file:linewithout revealing
the answer, lets you retry once, then reveals it with a short explanation. - 🏁 Score + weak areas —
Score: N/totalplus the topics to revisit, each
with a file pointer. Encouraging, not punitive. - 🌍 Your language — the quiz runs in whatever language you're writing in.
Demo

Installation
Via the Vercel skills CLI (recommended)
npx skills add arimet/handoff
This scans the repo's SKILL.md, detects your installed agent, and installshandoff into the right path (~/.claude/skills/ for Claude Code).
Manually
# Symlink (handy while developing — edits stay live)
ln -s "$(pwd)/skills/handoff" ~/.claude/skills/handoff
# …or copy
cp -r skills/handoff ~/.claude/skills/handoff
Full instructions — including enabling and tuning the optional hook — are in
skills/handoff/references/install.md.
Usage
After Claude finishes a feature, trigger it manually:
/handoff
…or just ask in natural language: "quiz me on what you built",
"vérifie que j'ai compris", "test my understanding", "comprehension check".
Auto-suggestion hook
An optional, non-blocking Stop hook nudges you to run /handoff when a
session produced a large diff. It only prints a suggestion — it never forces the
quiz and never blocks Claude. Register it in ~/.claude/settings.json:
{
"hooks": {
"Stop": [
{
"matcher": "",
"hooks": [
{
"type": "command",
"command": "$HOME/.claude/skills/handoff/scripts/suggest-quiz.sh",
"timeout": 5
}
]
}
]
}
}
Configuration
The hook suggests /handoff when either threshold is reached. Override the
defaults with environment variables:
| Variable | Default | Meaning |
|---|---|---|
CC_HANDOFF_LINES |
80 |
Min changed lines (added + removed) to suggest |
CC_HANDOFF_FILES |
4 |
Min changed files to suggest |
Counts include tracked changes (staged + unstaged) and new untracked files,
so files Claude just created are counted.
Requirements
- Claude Code with skills support.
- git (optional) — anchors questions on real
file:linelocations and
drives the hook. Without git, the skill still works from conversation context. - No extra tools to install. The hook parses its event with
node(always
present — Claude Code runs on Node) and falls back tojqif available. You
do not need to installjq.
Project layout
.
├── README.md
├── LICENSE
└── skills/
└── handoff/
├── SKILL.md # the skill: triggering + quiz procedure
├── scripts/
│ └── suggest-quiz.sh # optional, non-blocking Stop hook
├── references/
│ └── install.md # install + hook setup and thresholds
└── evals/
└── evals.json # test cases (excluded from the packaged .skill)
Packaging
Produce a distributable .skill archive (a zip with handoff/ at its root,
excluding evals/):
# Using Anthropic's skill-creator
python -m scripts.package_skill skills/handoff
Install the result by unzipping handoff.skill into ~/.claude/skills/.
Contributing
Issues and pull requests are welcome. When changing the skill:
- Keep
SKILL.mdlean — push detail intoreferences/. - Update
evals/evals.jsonwith a test case for new behavior. - Run
bash -n skills/handoff/scripts/suggest-quiz.shand smoke-test the hook.
License
Released under the MIT License.
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi