McDonalds Receipt Firecrawl

Reproducible Firecrawl workflow for McDonald's receipt survey navigation with non-submit safety mode, audit logging, and human takeover at final submission/captcha.

What this repository includes

• OpenClaw skill: skills/mcdonaldsreceipt/
• Deterministic runner script: skills/mcdonaldsreceipt/scripts/run_receipt_flow.py
• Prompt references: skills/mcdonaldsreceipt/references/prompts.md
• Output reports in JSON for reproducibility and debugging

Safety defaults

• Non-submit mode is the default behavior.
• The workflow is designed to stop before final submit/finish/complete.
• If a captcha/challenge appears at submission boundary, use human takeover.

Prerequisites

• Python 3.10+
• requests package
• Firecrawl API key

Install dependency:

pip install requests

Set API key:

export FIRECRAWL_API_KEY="<your-key>"

Quick start

One-command wrapper (recommended)

./run.sh "https://cxsurvey.sonata-software.com/ws/furl-mcdnl" "MQPF-KKRC-9JCM-WC46-CMYV" 8

Args: ./run.sh [url] [receipt-code] [audit-steps] [out-file]

Direct runner

python3 skills/mcdonaldsreceipt/scripts/run_receipt_flow.py \
  --url "https://cxsurvey.sonata-software.com/ws/furl-mcdnl" \
  --code "MQPF-KKRC-9JCM-WC46-CMYV" \
  --audit-steps 8 \
  --out "reports/mcdonaldsreceipt-$(date -u +%Y%m%d-%H%M%S).json"

Output format

• See reports/sample-report.json for a standardized example.

Each run saves JSON with:

• timestamp_utc
• target_url
• non_submit_mode
• scrape (success, scrape_id, content lengths, anti-bot flags)
• interact_1 (initial state discovery)
• interact_code (receipt code entry state)
• audit_logs (per-step question/selection trace)
• interact_2 (pre-submit summary)
• captcha_or_challenge_detected
• challenge_flags
• liveview_url

Human takeover flow (recommended)

Use this when submit may trigger captcha:

1. Run automation until pre-submit state.
2. Share liveview_url with the user/operator.
3. User takes over to solve any challenge and click final submit.
4. Store confirmation text/screenshot reference in your report.

For OpenClaw users

Skill entrypoint:

• skills/mcdonaldsreceipt/SKILL.md

Load that skill when you need to:

• test receipt survey navigation,
• collect anti-bot/captcha evidence,
• run repeatable, auditable non-submit flows,
• hand off final steps to humans.

Notes

• Do not hardcode API keys in committed code.
• Rotate keys if exposed.
• Firecrawl may return empty interact.output on some runs; rely on both output text and liveview_url evidence.