FoodLoop AI — Analyzer Skill

Reverse-engineers a FoodLoop AI deployment: maps routes, traces API endpoints, identifies the backend, and reconstructs the full user workflow.

Workflow

Step 1 — Identify Backend URL

Fetch the frontend's JS bundle and look for the backend host string:

curl -s "https://<frontend>/assets/$(curl -s 'https://<frontend>/' | grep -oP 'src="/assets/\K[^"]+')" | tr '"' '\n' | grep -E "https?://[^/]+\.(onrender|railway|vercel|render)" | sort -u

Or grep the JS for api/ route strings directly:

curl -s "https://<frontend>/assets/<bundle>.js" | tr '"' '\n' | grep -E "^/api/" | sort -u

Step 2 — Fetch OpenAPI Spec

Most FastAPI backends expose their spec at:

https://<backend>/openapi.json

This is the canonical source of truth — pull it first and extract all paths + schemas:

curl -s "https://<backend>/openapi.json" | python3 -c "
import sys, json
spec = json.load(sys.stdin)
for path, methods in spec['paths'].items():
    for method, details in methods.items():
        print(f'{method.upper():6} {path} — {details.get(\"summary\",\"?\")}')
for name, schema in spec.get('components',{}).get('schemas',{}).items():
    print(f'Schema: {name}: {list(schema.get(\"properties\",{}).keys())}')
"

Step 3 — Map Pages

Probe common Next.js routes:

for route in "" "home" "scan" "analyze" "result" "dashboard" "about" "how-it-works" "login" "signup"; do
  curl -s -o /dev/null -w "$route → HTTP %{http_code}\n" "https://<frontend>/$route"
done

Step 4 — Trace the User Workflow

Use the OpenAPI spec as reference. The core workflow is:

1. Input → POST /api/analysis/text or POST /api/analysis/upload
2. Detect → AI identifies leftovers (label, display_name, confidence, safety_flag)
3. Clarify (optional) → POST /api/analysis/{id}/clarify
4. Recommend → POST /api/recipes/recommend/{analysis_id} → Indonesian recipe list
5. Cook → POST /api/recipes/sessions/{id}/start → guided session
6. Track → POST /progress, POST /finish, POST /stop
7. Remind → Notifications via GET /api/notifications/preview
8. Analyze → Dashboard at /api/dashboard/summary
9. Export → POST /api/pdf/generate/{recommendation_id}

Step 5 — Test the API

Test demo auth:

curl -s -X POST "https://<backend>/api/auth/demo" \
  -H "Content-Type: application/json" \
  -d '{"name":"Test User","email":"test@example.com"}'

Test text analysis:

curl -s -X POST "https://<backend>/api/analysis/text" \
  -H "Content-Type: application/json" \
  -d '{"text":"nasi sisa, telur, sayuran","condition":"segar"}'

Step 6 — Check Config

curl -s "https://<backend>/api/config/status"

Step 7 — Document Findings

Save the complete analysis to: ~/.openclaw/workspace/foodloop_<domain>_<date>.md

Include: architecture table, all endpoints, workflow steps, test results, and schema summaries.

Analyze a Photo (from chat)

When the user sends a photo directly in chat, analyze it using the upload endpoint:

Step 1 — Save the photo to a temp file:

# If image is a base64 data URL (most common from chat):
python3 -c "
import sys, base64, re
data = sys.stdin.read()
# Handle data URL or raw base64
if data.startswith('data:'):
    header, data = data.split(',', 1)
img_bytes = base64.b64decode(data)
ext = 'jpg'
if 'png' in header.lower(): ext = 'png'
path = '/tmp/foodloop_upload.jpg'
with open(path, 'wb') as f:
    f.write(img_bytes)
print(path)
" <<< "$(cat /path/to/image_data_or_data_url)"

# Or if the photo is already saved to a known path, use it directly

Step 2 — Upload and analyze the photo:

curl -s -X POST "https://<backend>/api/analysis/upload" \
  -F "file=@/tmp/foodloop_upload.jpg" \
  -F "condition=segar"

The condition field is optional. Common values: segar (fresh), ragu (uncertain).

Step 3 — Parse and present the results:

curl -s "https://<backend>/api/analysis/{analysis_id}"

Output format — same AnalysisResponse as text analysis:

• items[] — detected leftovers with label, display_name, confidence, safety_flag
• safety_level — overall safety assessment
• warnings[] — food safety notes
• recommendations[] — Indonesian recipe suggestions with score, reason, ingredients, steps

Supported image formats: JPEG, PNG, WebP (detected automatically by backend from Content-Type)

Key Endpoints (FastAPI/FoodLoop AI Backend)

Backend Detection Pattern

FoodLoop AI backends expose a FastAPI doc UI at /docs. If /openapi.json returns a valid spec, confirm it contains "FoodLoop AI" in the info.title field.

Reference

For the complete API schema, endpoint parameters, request/response schemas, and live test examples, see references/api_workflow.md.