ask-ladder

socratic tutor-prompt scaffolding misconceptions k12
by @iberry420Education
Build a paste-ready Socratic tutor prompt for one learning objective: question repertoire, misconception branches, stall scaffolds, escape hatch, safety/scope stops, and a teacher companion with a 4-turn good/bad sample. Use when a teacher wants a Socratic AI tutor, ask-dont-tell prompt, misconception branches, or student AI study tool that questions instead of answers. Education only.
SKILL PACKAGE

Directory layout for Grok — SKILL.md plus scripts, references, and assets. 6 file(s).

Select a file

Edit in place, then Save (full package security scan). Use fullscreen for a larger workspace.

SYNCED SUMMARY (from SKILL.md)
# Ask Ladder

Hold the tutor in **question mode**. For **one** learning objective, produce a paste-ready **tutor prompt** (question repertoire, misconception if-thens, 3-step stall scaffold, reflective pauses, escape hatch, safety/scope stops, no-PII, end-of-session summary) plus a **teacher companion** and a **4-turn fits / does-not-fit** sample dialogue.

Inspired by KRASA-AI/education-ai-skills socratic-tutor-prompt-builder (MIT). Copyright (c) 2026 KRASA AI (inspiration only; credited, not copied). Original English Grok remake. Complements study-loop (flashcards/quiz from notes), glow-grow (feedback on finished work), pulse-check (quick CFU), claim-line (essay spine), prep-note (teacher lesson internalization). Education only — formative tutor prompt craft, not therapy, not grading, not a companion persona.

Read `references/ask-rails.md` before you write. Fill the three templates under `templates/`.

## When to use

- Teacher wants a **Socratic AI tutor prompt** for one objective (math step, claim-evidence, conjugation, stoichiometry, thesis revision)
- Retrofit a generic chatbot into ask-dont-tell mode for a workstation / rotation
- Need misconception branches, stall scaffolds, escape hatch, and safety routing baked into the prompt
- Need a teacher-facing companion + sample dialogue to judge whether the deployed prompt behaves

## When NOT to use

- Direct-instruction explainer that should answer first → not this skill
- Graded assessment / quiz bank as the main job → quiz-bank or pulse-check
- Essay outline / thesis spine without a live tutor prompt → claim-line
- Study cards / spaced quiz from notes → study-loop
- Feedback comments on finished student work → glow-grow
- Therapy, counseling, SEL crisis coaching as the product (route to human counselor; safety stops only)
- Companion / persona / character chatbot for students (especially under 16) — tutor must stay an AI study tool
- Doing the student's homework, writing their essay, or leaking summative keys

## Inputs

Gather what you can. Label gaps `[GUESS]` for the teacher to confirm. Ask at most two questions if subject/grade or the single objective is missing, then proceed. **Never invent real student PII.**

- Subject + grade band (required)
- **One** learning objective (required) — multiple objectives → split into separate prompts
- Student entry point (what a typical first attempt looks like)
- 2–3 predictable misconceptions (wrong-but-reasonable)
- Approved deployment surface (district tool name, or "generic tutor prompt")
- Expected turn budget (3–5 / 6–10 / 10+)
- Scope hard-stops (no essay writing, no summative keys, no grading, …)
- Escape-hatch policy (default: two consecutive "I'm stuck" → one analogous worked example → back to questions)
- Teacher transcript review cadence if known (real-time / daily / weekly / unknown)
- Named classroom scaffolds the prompt may point to (anchor chart, sentence frames) — names only
- Safety contact role label for routing (e.g. "school counselor") — **do not invent a real person's full contact details**

## Procedure

1. Read `references/ask-rails.md`. Refuse therapy-as-product, companion personas for minors, grading engines, homework-completion bots, live exam keys, extra PII, CSAM. Crisis language → fixed safety stop + US 988 if self-harm appears in the *authoring* chat, then stop.
2. Fill `templates/objective-card.md`. One objective, entry point, misconceptions, surface, turn budget, scope stops, escape hatch, review cadence, scaffolds, disclosure needs.
3. Fill `templates/tutor-prompt.md`. Paste-ready prompt: role first, question ≥80%, identity-as-AI, repertoire, misconception branches, 3-step scaffold, reflective pause every 3–4 turns, escape hatch, verbatim safety stop, scope stop, no-PII, end summary. Open-with student line.
4. Fill `templates/teacher-companion.md`. Scope statement, deployment notes, escape/safety/scope lists, review cadence, retire-and-replace conditions, 4-turn fits vs does-not-fit table. Bracket missing school config as `[TEACHER FILL]`.
5. Keep examples original and age-appropriate. Do not copy living commercial tutor prompt dumps wholesale.
6. Deliver the three templates. Closing line: ask-ladder tutor-prompt worksheet; formative only — not therapy, not a grader, not a persona.

## Output format

- Objective card (one goal, misconceptions, stops, hatch)
- Tutor prompt + open-with line (paste-ready)
- Teacher companion + 4-turn good/bad sample
- One line: ask-ladder worksheet; not therapy and not a companion persona

## Safety

- Follow `references/ask-rails.md` with no exceptions.
- Education formative design only.
- Tutor presents as an **AI study tool**, never a human or named companion character for students under 16.
- Safety stops for self-harm / abuse / weapons / crisis are **verbatim** and end the student session.
- No PII collection beyond optional first name if the platform supplies it.
- No CSAM. No writing graded work for the student. No leaking live summative keys.
- If the *teacher* chat itself describes an active crisis, give US 988 / local emergency guidance and stop building scenario details.

## Stop conditions

- Three templates delivered, or refused for therapy-as-product / companion-persona-for-minors / grader-as-main / homework-bot / live keys / extra PII / CSAM / study-loop-or-glow-grow-as-better-fit
Version History
Comments (0)
No comments yet. Be the first!
Sign in to leave a comment.