/teach-test

Make it your own.

In Claude Code, use the slash command and add your context. In Codex, select teach-test from the just-vibe skill picker, then send the same brief.

Version 0.11.0 also supports /jv teach-test, /just-vibe teach-test and /jv:teach-test in Claude. See shortcut setup and context examples.

What the agent does

1. Read ../../references/teaching.md and ../../references/teach-test.md. Resolve the topic from the complete brief or immediately preceding lesson; ask for it only if unresolved.
2. Inspect actual callable tools. Prefer request_user_input_async in a Codex environment that exposes and permits it, otherwise request_user_input only in its supported mode; use AskUserQuestion in Claude when available. Follow the actual host schema and restrictions. If no suitable dialog is available, explain that native interactive testing is unavailable here and stop without inline questions.
3. Create quiz state with toolkit quiz create. Generate a question grounded in the lesson or target workflow, with exactly three short neutral choices, a correctOptionId, concept, and explanation. Keep this answer material out of the user-facing question.
4. Call toolkit quiz present to validate and shuffle the choices and produce a presentation payload for the observed dialog. Actually invoke that native host tool with presentation.arguments. Merely printing the payload does not ask the question.
5. For asynchronous tools, retain the quiz as awaiting-answer and wait for a user response; a returned pending tool call or preselected option is not submission. Do not advance, reveal the answer, or end with an inline replacement question while waiting.
6. Normalize the actual response into questionId, submitted, selection/freeText, skipped/cancelled and call toolkit quiz answer. For a free-text explanation, use quiz review only after assessing its meaning; preserve the actual response and explain the judgment.
7. Show returned feedback after answers in practice mode; withhold correctness in test mode until finished. Generate the next question using demonstrated misconceptions and remaining budget, then repeat through the native tool. Use quiz report for the final assessment; report sample limits and optional review topics.
8. Test one concept at a time, mixing a state prediction with an application when useful. Adapt to the specific misconception; a short quiz cannot establish mastery.

Inputs

• Topic or recent lesson and verified access to an appropriate native multiple-choice user-question tool.

Optional context: questionCount, experienceLevel, difficulty, mode, language, references.

Scope

Reads

Assess understanding through one native dialog question at a time. Do not render quiz questions inline, implement the named tool, alter project files, or change host mode/settings automatically.

Writes

No project or external-service writes. Keep quiz state in session context; on an explicit save request, write the quiz report to the path the user names or save it with project checkpoint NAME.

Mode

Inspect; a topic, workflow, or recent lesson plus optional experience, question count and practice/test preference. Requires an available native question tool permitted for this interaction in the current host mode.

Prerequisites

Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.

Expected output

• Native multiple-choice question dialogs, answer-dependent feedback, and a final short assessment with concepts to revisit. Question text stays in the dialog; explanations may appear in normal conversation after answers.
• Submitted-answer evidence, grading rationale, skips, and a scoped assessment of demonstrated knowledge.

How the work is checked

• The host receives question/options only, with no answer key or explanation in the native payload. A second question cannot be presented before the pending answer is resolved.
• Empty, absent, default-only and delayed async responses do not advance or grade the quiz. Skips/cancellations do not count as incorrect; stale answers cannot answer a newer question.
• A linked-list quiz checks traversal, known-node insertion and pointer order. An ml-split quiz checks temporal/entity leakage and never starts training. If native dialogs are unavailable, the workflow reports that limitation and does not substitute inline quiz questions.

When to stop or clarify

• Stop on the requested question limit, cancellation, missing native-dialog capability, ambiguous topic, or an unresolved answer. Respect tool-specific host/mode restrictions; do not force Plan mode, grant permissions, or simulate a native dialog by printing JSON or markdown.

Handling missing context

Infer

Resolve the named files, existing scripts, current task and earlier corrections from the conversation and repository.

Assume

Use the narrowest interpretation that completes a reversible local task; state a consequential assumption once.

Ask

Ask when competing targets or incompatible success conditions would change the result; continue independent inspection first.

Technical guidance

Evidence

Inspect requested topic, level, previous answers and availability of the host's native question dialog.

Method

Ask one discriminating multiple-choice question in the dialog, wait for the answer, explain why choices differ and adapt the next question.

Pitfall

Inline questions or an early answer key defeat native assessment; if the host forbids quiz dialogs, explain the limit and stop the quiz without disguising it as clarification.

Check

Verify that questions wait for a real answer and that a misconception changes subsequent practice without claiming mastery from one guess.

Situational decisions

When question tool is missing or restricted to clarification: Report that assessment is unavailable in this mode; do not disguise quiz questions as implementation clarifications.

The coding agent follows this workflow using its available tools. Installation does not grant service access or guarantee an outcome. Read the compatibility notes.

Keep exploring