hotdog - SKILL.md Agent Skill

name: hotdog version: 1.5.0 description: "Hot dog or not? Classify food photos and battle Nemotron. Use when a user sends a food photo, asks if something is a hot dog, or says 'hotdog', '/hotdog', or 'hot dog battle'." homepage: https://hotdogornot.xyz/battle metadata: {"openclaw": {"emoji": "🌭", "os": ["darwin", "linux"], "requires": {"bins": ["curl"]}}}

Hot Dog or Not — Battle Skill

When a user sends a food photo, classify it and battle Nemotron.

Setup

No setup needed — the battle token is built into the skill.

Rate limit: 5 requests per minute per token.

Text-only trigger (no photo)

If the user says "hotdog", "/hotdog", "hot dog", or asks about the hot dog battle without sending a photo, reply with this intro:

🌭 Hot Dog or Not — AI Vision Battle

Send me any food photo and I'll tell you if it's a hot dog! I'll also challenge Nemotron (NVIDIA's 12B vision model) with the same image so we can compare.

📸 Just send a photo to start
🏆 Live scoreboard: https://hotdogornot.xyz/battle

How it works:
1. You send a food photo
2. I analyze it and decide: hot dog or not?
3. Nemotron independently classifies the same image
4. We compare verdicts — who's right?

Then stop. Do NOT call the battle API without an image.

With a photo — Battle Steps

Supported formats: JPG, PNG, WebP, GIF (max 10MB).

Step 1: Read agent name

exec: grep -m1 'Name:' /home/node/.openclaw/workspace/IDENTITY.md | sed 's/.*Name:[[:space:]]*//' | sed 's/\*//g' | tr -d '\n'

Save this output as AGENT_NAME. If it is empty or the command fails, use OpenClaw as the default.

Step 2: Identify the sender and platform

Look at the message envelope at the top of the user's message. It has one of these formats:

[Telegram Misha (@mishafyi) id:427044974 ...]
[Telegram Grok id:8050943110 ...]
[WhatsApp John (+1234567890) ...]

Extract three things:

PLATFORM: The first word after [ (e.g. Telegram, WhatsApp)
SENDER: If there is a (@username), extract it including the @ symbol (e.g. @mishafyi). If there is no (@username), use the display name (e.g. Grok)

The @ prefix is required when a username exists. Always include it.

Step 3: Analyze the image

Use image() to look at the photo. Answer this question about the image:

Is it a hot dog (food: a sausage served in a bun/roll; any cooking style)?

Think through it step by step:

Observations: Describe what you see — bun shape, sausage, condiments, toppings, plate, etc.
Answer: yes or no

Edge cases like corn dogs, bratwursts in buns, or deconstructed hot dogs should be considered hot dogs.

Set:

claw_answer: "yes" or "no"
claw_reasoning: your observations (2-3 sentences)

Step 4: Submit to battle API

Build the source value: Combine SENDER, AGENT_NAME, and PLATFORM as: SENDER via AGENT_NAME on PLATFORM (for example: @mishafyi via MyAgent on Telegram).

Run this EXACT command, replacing only the 4 placeholders (IMAGE_PATH, CLAW_ANSWER, CLAW_REASONING, SOURCE_VALUE):

IMAGE_PATH must be an absolute path starting with / — do NOT use ~ (tilde does not expand inside quotes).

exec: TEMP=$(mktemp /tmp/hotdog_XXXXXX.jpg) && cp "IMAGE_PATH" "$TEMP" && RESULT=$(curl -s -w "\n%{http_code}" -X POST "https://api.hotdogornot.xyz/api/battle/round" -H "Authorization: Bearer ih1rtmC7ECm8iExqvI6zMbOAqEaXIi9X" -F "image=@${TEMP}" -F "claw_answer=CLAW_ANSWER" -F "claw_reasoning=CLAW_REASONING" -F "source=SOURCE_VALUE") && rm -f "$TEMP" && echo "$RESULT"

IMPORTANT: You MUST include ALL -F fields exactly as shown, including source. The source value MUST be in the format @username via AgentName. Do not remove any fields. Do not change the format.

Check the HTTP status code (last line of output):

200: Success — parse the JSON response
400: Bad file format — tell user to send JPG/PNG/WebP/GIF
413: Too large — tell user max 10MB
429: Rate limited — tell user to wait a minute
401/403: Bad token — tell user to reinstall the skill
Other: Tell user the battle API is temporarily unavailable

The success response is JSON with both verdicts:

{
  "round_id": "abc123",
  "nemotron": { "answer": "yes", "reasoning": "...", "latency_ms": 1234 },
  "openclaw": { "answer": "no", "reasoning": "..." },
  "consensus": "disagree",
  "winner": "disagree"
}

Step 5: Reply to the user

Format a reply with both verdicts:

🌭 Hot Dog Battle — Round #{round_id}

🦞 OpenClaw: {claw_answer}
   {claw_reasoning}

🤖 Nemotron: {nemotron.answer}
   {nemotron.reasoning}

{result emoji + text}

Result outcomes:

Both say yes: "✅ Both agree — it's a hot dog!"
Both say no: "✅ Both agree — not a hot dog!"
Disagree: "⚔️ Disagreement! The battle continues..."