Then branch on the status:

**`FOUND_REASONING`** (o3, o1, DeepThink, DeepSeek R1) — best case, use for problem generation:

No warning needed. Sub-agent writes the .tex files + verify script, announces completion.

**`FOUND_STRONG`** (Claude Opus) — excellent quality, use it without alarming the user:

Optionally add a quiet note: *"Using Opus — solid math accuracy and excellent LaTeX. For the hardest Algebra 2 problems, a reasoning model (DeepThink/o1) would be marginally better."*

**`NONE`** — standard model only; proceed but surface a clear recommendation:

| Status | Model examples | Action |
|---|---|---|
| `FOUND_REASONING` | DeepThink, o1, o3, R1 | Spawn silently, no warning |
| `FOUND_STRONG` | Claude Opus 4.x | Spawn silently, optional quiet note |
| `NONE` | Sonnet, Flash, GPT-4o | Use current model + show recommendation |

## Prerequisites

Output directory (create if needed): `~/Documents/Worksheets/`

## Workflow

### 1. Gather requirements

Ask (or infer from context):
- **Student**: name, grade, course (e.g. "8th grade, Pre-Algebra")
- **Topic**: e.g. "factoring trinomials", "solving two-step equations"
- **Problem count**: default 10 if not specified
- **Format preference**: timed quiz, homework practice, mixed difficulty, or topic drill

**Photo input shortcut**: If the user sends a photo of homework or a textbook page, use the `image` tool to extract problem types, format, and difficulty — then mirror that style exactly.

### 2. Design problems

Design problems appropriate to the student's level. Increase difficulty gradually across the set. Every problem must be mathematically correct — verify your own solutions.

See `references/problem-library.md` for topic-specific problem type menus.

### 3. Write LaTeX source

Write **three** `.tex` files to `/tmp/`:
- `ws_TOPIC_DATE.tex` — student worksheet (blank work areas)
- `ak_TOPIC_DATE.tex` — answer key (full step-by-step solutions)
- `ss_TOPIC_DATE.tex` — skills summary / study guide (cheat sheet)

The **skills summary** is a 1–2 page reference card the student can use while working through the worksheet or when studying. It contains:
- One section per distinct skill tested (2–5 sections typical)
- A **formula/rule box** (blue) per skill — the key facts and formulas
- A **mini worked example** (green) per skill — brief pattern demonstration, simpler than worksheet problems
- An optional **watch-out box** (orange) — common mistakes worth flagging
- Optional **key vocabulary** section at the bottom

See `references/latex-templates.md` → "Skills Summary / Study Guide Template" for the full shell and box macros.

See `references/latex-templates.md` for document templates, coordinate planes, tables, geometric figures, and answer key patterns.

**Required packages** (include in every document):

**Work space defaults**: `\vspace{5cm}` per problem; `8cm` for multi-step; `10cm+` for graphs.

### 4. Write and run the verification file

Before compiling, write `/tmp/verify_TOPIC_DATE.json` — a structured data file describing each problem and its expected answer. The bundled `scripts/verify.py` evaluates this using SymPy. No generated code is ever executed.

**Type reference:**

| Type | Verifiable? | What it checks |
|---|---|---|
| `solve` | ✅ | Roots of expr=0 match expected list |
| `factor` | ✅ | Factored form matches expected |
| `expand` | ✅ | Expanded form matches expected |
| `eval` | ✅ | expr evaluated at given values matches expected |
| `zeros` | ✅ | Zeros of expr match expected list |
| `manual` | 👁 | Flagged for human review — never fails automatically |

Use `manual` for: graph sketches, sign charts, word problem setups, proofs.

**If verification fails (exit 1):** fix the LaTeX answer key and re-run. Do not compile until the answer key is correct.

### 5. Compile

### 5. Deliver

Send all three PDFs via the same channel the request came from:
- **Telegram** → `message` tool with `filePath` (copy to `~/.openclaw/media/outbound/` first)
- **iMessage/SMS** → `imsg` skill
- **Email** → `gog` skill (send all three as attachments)

Suggested send order: skills summary first (study guide), then worksheet, then answer key.

**Printing**: Do NOT print unless explicitly asked. If asked, print worksheet + skills summary (not answer key, unless requested). Use `lpr -P <printer_name>`.

## Quality Checklist

Before compiling, verify each problem:
- [ ] Mathematically correct (you checked the solution)
- [ ] Unambiguous problem statement
- [ ] Appropriate difficulty for the student's level
- [ ] Sufficient work space
- [ ] Diagrams/graphs/tables included where needed
- [ ] Problems vary across the set (not all the same sub-type)

## File Naming

SKILL_DIR="$(dirname "$0")"
result=$(bash "$SKILL_DIR/scripts/check_reasoning_model.sh")
status=$(echo "$result" | awk '{print $1}')   # FOUND, FALLBACK, or NONE
model_alias=$(echo "$result" | awk '{print $2}')
model_full=$(echo "$result" | awk '{print $3}')

⚠️ No reasoning model or Opus detected. Worksheet generated with [current model].
For best accuracy, especially on multi-step problems, configure one of:
  • Gemini 2.5 Pro DeepThink  — google.generativeai.com (free tier available)
  • o1 / o3                   — platform.openai.com
  • DeepSeek R1               — platform.deepseek.com (very affordable)
  • Claude Opus               — console.anthropic.com
SymPy verification will catch most errors regardless.