agentleFS
dalvgitdalvgit/vibecheck/skills/exam/SKILL.md

exam

Skill4 starsChanged 4 months agoMITView on GitHub

Examine the human on the code they are about to push. Use when a push is blocked by vibecheck, or when the user asks to be vibechecked. Reads the outgoing diff, asks 2-4 multiple-choice questions about it via AskUserQuestion, grades them, and unlocks the push on a pass.

6 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions6 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 6

    1. dull-bird/teach-me-skill★ 2same name
    2. dymchenkko/claude-study-agent★ 2same name
    3. LazyerIJ/self-clone★ 2same name
    4. devinat1/skills★ 1same name
    5. jiangtao/skills★ 1same name
    6. kubilaiswf/claude-tutor★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every blocked push prints the gate's absolute path on its last line.

    line 24
  2. Very specific:

    Read .vibecheck.toml if it exists: questions (default 3) and pass (all | majority, default all).

    line 42
  3. Very specific:

    Follow "$GATE" --guide. In short: find the 2–4 *ideas* the change rests on and test what follows from them.

    line 47
  4. Very specific:

    Ask all questions in a single AskUserQuestion call — one per questions entry.

    line 53
  5. Distinctive:

    Do not reveal the correct answer, hint at it, or react to answers until everything is in.

    line 67
  6. Distinctive:

    Never answer for the human, and never issue a receipt they didn't earn.

    line 83

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

vibecheck: the exam (Claude Code)

An agent wrote this code; the human is about to put their name on it. Your job is to find out whether they understand what they're shipping. You are the examiner, not the candidate — never answer, hint, or narrow the options. AskUserQuestion is the only way to collect an answer here.

This file is only the Claude Code delivery wrapper. The instructions that decide whether the exam is any good — what to ask, the altitude, fair distractors, how to grade, the one rule — live in the shared guide. Read it before writing questions and follow it:

"$GATE" --guide

0. Find the gate

Every blocked push prints the gate's absolute path on its last line. Use that. If you're run without a block message (the user just asked to be vibechecked), try "${CLAUDE_PLUGIN_ROOT}/hooks/gate.sh", and failing that:

ls ~/.claude/plugins/*/vibecheck*/hooks/gate.sh 2>/dev/null || find ~/.claude -name gate.sh -path '*vibecheck*' 2>/dev/null | head -1

1. Read the diff

"$GATE" --diff

This is exactly the diff the gate hashes — do not substitute your own git diff, or the receipt won't match and the push stays blocked. If it's empty, say so and stop. Use Grep to read surrounding code (callers, etc.) where a question needs it.

Read .vibecheck.toml if it exists: questions (default 3) and pass (all | majority, default all).

2. Write the questions

Follow "$GATE" --guide. In short: find the 2–4 ideas the change rests on and test what follows from them; state the change plainly and make them reason out the consequence; every distractor actually wrong and distinguishable; never quiz absent code or line-level trivia.

3. Administer via AskUserQuestion

Ask all questions in a single AskUserQuestion call — one per questions entry. The human answers; you cannot.

Every question carries the way out. Make the last option of every question a plain, first-class escape — same wording each time:

  • …(the substantive answers)…
  • Skip the vibecheck — push without answering

That leaves three substantive options per question (one correct, two wrong — the guide's rules apply to those). The point: the human can bail at the very first question, or any question, without reading the rest. If they pick the skip option on any question, that is a skip — honor it immediately (--skip) and do not grade.

Do not reveal the correct answer, hint at it, or react to answers until everything is in.

4. Grade and record

Score per pass (all = every question; majority = more than half). For how to handle a fail, a skip, or pushback, follow the guide — be helpful on a fail, take a skip at face value, believe them when a question was unfair.

  • Pass: "$GATE" --pass, then retry the push and report the score plainly.
  • Skip (picked from the exam, or said at any point): "$GATE" --skip, immediately, no guilt-tripping. Then retry the push.
  • Fail: issue no receipt; explain what the code actually does, name the hunks worth reading, and offer the three doors (try again / skip / go read).

The one rule

Never answer for the human, and never issue a receipt they didn't earn. You could forge it — you have a shell — which is exactly why it matters that you don't. See the guide's closing lines.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.