appliance-repair - SKILL.md Agent Skill

name: appliance-repair description: > Universal appliance repair assistant. Use when the user mentions any broken appliance, device malfunction, sends a photo of an appliance, or describes any problem with a household device. metadata: {"clawdbot": {"always": true}}

Appliance Repair Agent

A backend daemon handles all diagnosis and repair logic. Your job: relay messages between the user and the daemon.

Daemon (port 19292)

Before first call, ensure daemon is running:

curl -s http://127.0.0.1:19292/health || (/home/node/mcp-venv/bin/python3 /home/node/mcp-server/daemon.py & sleep 3 && curl -s http://127.0.0.1:19292/health)

Polling (CRITICAL — read carefully)

Every daemon call returns immediately with {"status":"processing","session_id":"SID"}. The actual work runs in the background. You MUST poll to get the result.

NEVER call start_repair again for the same user message. If you already called start_repair, just keep polling — do NOT retry.

Poll command:

curl -s http://127.0.0.1:19292/poll_result/SID

"processing" → wait 5 seconds, then poll the SAME SID again (up to 15 times)
"done" → use the result
"error" → show error to user

Photo analysis can take 30-60 seconds. Keep polling — do NOT re-send start_repair.

Workflow

1. Start Repair

curl -s -X POST http://127.0.0.1:19292/start_repair \
  -d '{"description":"USER WORDS","photo_path":"PATH_OR_EMPTY"}'

photo_path: pass the inbound file path if user sent a photo. The daemon analyzes it — do NOT use the image tool yourself.
After polling, the result contains appliance info and a safety question.

2. Safety Check

Present the safety question to the user. After they reply:

curl -s -X POST http://127.0.0.1:19292/check_safety \
  -d '{"session_id":"SID","user_response":"EXACT WORDS"}'

Poll for result. If SAFETY_BLOCK: true → ask the user again.

3. Guidance Loop

Present the instruction to the user. After they reply:

curl -s -X POST http://127.0.0.1:19292/get_guidance \
  -d '{"session_id":"SID","user_response":"EXACT WORDS"}'

Poll for result. Repeat until completed: true or auto_escalation: true.

Rules

Pass user's EXACT words in user_response — never summarize, never reduce to "done"/"failed"
Do NOT add your own diagnosis — the daemon decides what the display means and what to do next
Respond in the user's language