Catering Purchase Allocation Check
Find possible duplicate entries in your event receipts and restaurant purchases.
Restaurant purchasing
Recheck the figures in your agreement and eligible invoices against the rules you supply.
A downloadable skill for your AI agent.
Documented progress against supplied thresholds.
Source references, missing information, and questions to review stay with the findings.
Open your download and follow the start guide. Use an AI agent that can load skill files and run the included tools.
Agreement and eligible invoices. Confirm the case and period you want to check.
Copy the request below into your agent. Add the files and any task details it needs.
Check the findings against the original records. Choose where to save the report and follow up on open questions.
Use the Restaurant Rebate Threshold Worksheet skill. Review the attached records: agreement and eligible invoices. Prepare this result: documented progress against supplied thresholds. Include a source reference for each finding. List missing information and questions that need my review. Ask me to confirm the scope and where to save the report before you start.
It records the rule and the values used in each calculation. It works out the expected amount and compares it with the reported amount.
Agreement and eligible invoices.
For this task, keep one review row for each rebate agreement + period. Record eligible purchases, threshold, rate, and exclusions. Keep the source location beside each value.
Normalize purchase quantities using sourced pack size and units. Keep price, credit, free goods, substitution, waste, and inter-location transfer evidence distinct.
Compare like-for-like products and dates. A different pack can hide a unit-price increase. Use approved ingredient and food-handling records.
The helper validates arithmetic, not the legal or business rule. It does not infer a formula, evaluate spreadsheet functions, or round silently. Use a sourced tolerance for rounding comparisons.
This synthetic example shows the included check in use. Your agent uses its findings to complete your task report.
Synthetic example records
| Sample record | expression: quantity * rate; variables: quantity: 4; rate: 25.00; observed: 101.00; unit: example units |
|---|
A difference needs review
This result comes from the checking tool in your download.
These are synthetic records from the bundled example.
Synthetic rule: amount = quantity * rate; no tolerance. Not customer evidence.{"expression": "quantity * rate", "id": "r1", "key": "rebate agreement + period / SYNTHETIC-001", "observed": "101.00", "rule_source_id": "rule", "unit": "synthetic-unit", "variables": {"quantity": "4", "rate": "25.00"}}After purchase, extract the ZIP. Run these commands from the skill folder. Use a new output folder.
python3 scripts/checks.py
python3 scripts/audit.py assets/example.json \
--files-root assets/sample-files --output-dir example-result
Expected exit code: 1. Code 0 means no helper finding. Code 1 means a finding or review item. Code 2 means incomplete or invalid input.
The check uses the records you supply. Review each finding against the original documents.
Ask your agent to save the completed task report in a folder you choose. The report template includes the result, source references, missing information, and next actions.
Tonotes does not receive your work documents through this website. Documents you share with an AI service follow that service's storage settings.
US$11 includes one skill in a ZIP with 18 files. You receive the instructions and tools to run the task with your own records.
Use an agent that supports SKILL.md instructions and Python 3.10 or later with SQLite. Your AI-agent service is separate.
Start with agreement and eligible invoices. Confirm the case, period, and source versions before the review.
Use supplier, product, pack specification, invoice line, location to keep records distinct. The agent records the original value, an exact quote, and the source location.
Apply the supplied formula and compare the calculated amount with the reported amount. Run the Python helper on the structured records. Check each finding against its source.
Prepare the task result: documented progress against supplied thresholds. Include missing evidence, unresolved questions, and the next action for each open item.
Eight synthetic cases check the tool locally. Live-agent accuracy on real documents has not been tested.