handoff

agent
Guvenlik Denetimi
Gecti
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.

SUMMARY

Claude Code skill that quizzes you on the code Claude wrote — verify you understood it before shipping.

README.md

🤝 handoff

Verify you actually understood the code Claude wrote — before you ship it.

Claude Code
Hook
Zero install
License: MIT


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

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) with git diff (the where: real
    file:line locations), 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:line without revealing
    the answer, lets you retry once, then reveals it with a short explanation.
  • 🏁 Score + weak areasScore: N/total plus 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

handoff comprehension check 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 installs
handoff 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:line locations 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 to jq if available. You
    do not need to install jq.

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:

  1. Keep SKILL.md lean — push detail into references/.
  2. Update evals/evals.json with a test case for new behavior.
  3. Run bash -n skills/handoff/scripts/suggest-quiz.sh and smoke-test the hook.

License

Released under the MIT License.

Yorumlar (0)

Sonuc bulunamadi