Receipts to Expense Ledger
Turn a folder of receipt photos/PDFs into a clean expense ledger: OCR each receipt, extract vendor, date, line items, tax, and total, assign an expense category, and reconcile that each extracted total matches its line items. Locks the per-receipt field schema, the category set, and the math/reconciliation rules FIRST, then OCRs and extracts each receipt into rows, then verifies every receipt was processed, totals reconcile, currencies/dates are normalized, and low-confidence reads are flagged for review. Defining the schema and the total-equals-line-items check before extraction is what keeps an expense ledger trustworthy for reimbursement or tax.
Steps
Entry step: scope. Each step names the specialist role it wants; the full working prompt is expandable.
- Scope the ledgerplannerentry
per-receipt schema, categories, reconciliation rules
Show working prompt
Define the ledger contract before reading receipts. Step 1: identify the receipts folder and the file types (jpg, png, pdf). Step 2: define the per-receipt schema (file, vendor, date, currency, line items[{desc, amount}], subtotal, tax, total, category, confidence, needs_review). Step 3: define the closed expense-category set (e.g. Meals, Travel, Supplies, Software, Other) — every receipt gets exactly one. Step 4: define the math/reconciliation rule — sum(line items) + tax should equal total within a small tolerance; a mismatch sets needs_review. Step 5: define normalization (dates → ISO, amounts numeric to two decimals, a single output currency or a currency column) and the confidence threshold below which a row is flagged. Step 6: write an acceptance-criteria checklist ('every receipt file produces a row', 'amounts numeric and dates ISO', 'total reconciles to line items + tax within tolerance else flagged', 'category is in-set', 'low-confidence reads flagged for review', 'a grand total across receipts is reported'). Call write_task_note with schema + categories + rules + checklist and write notes/scope.md. No OCR yet. - OCR and extractdeveloper
read each receipt, extract fields, categorize, reconcile
Show working prompt
Process every receipt into a ledger row to the locked schema. Step 1: list the folder and, for each receipt, use the available OCR/vision capability to read the text. Step 2: extract vendor, date, line items, subtotal, tax, and total; normalize dates to ISO and amounts to numbers. Step 3: assign exactly one in-set category from the vendor/items. Step 4: reconcile — check sum(line items)+tax ≈ total within tolerance; on mismatch or low OCR confidence, set needs_review with a note rather than silently trusting it. Step 5: write all rows to the output file (JSON), confirm it parses, and compute a grand total across receipts. Do not skip any receipt; if one is unreadable, still emit a row flagged needs_review. Call write_task_note with the output path, the receipt count, and how many are flagged.
- Verify the ledgerreviewer
coverage, reconciliation, normalization, review flags
Show working prompt
Verify the ledger against scope. Step 1: confirm the file parses and there is one row per receipt file in the folder — none skipped. Step 2: confirm amounts are numeric to two decimals and dates are ISO. Step 3: confirm reconciliation ran — rows where line items + tax do not match the total within tolerance are flagged needs_review, and clean rows actually balance (re-check 3 by hand). Step 4: confirm categories are all in-set and low-confidence/unreadable receipts are flagged rather than silently accepted. Step 5: confirm the grand total equals the sum of the row totals. Write PASS/FAIL per criterion to notes/verify.md and task notes; on failure, name the receipts and loop back to extract.
- Evaluatereviewer
Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.
Show working prompt
Open ledger.json and verify every criterion from notes/scope.md: one row per receipt file (none skipped, unreadable ones flagged), amounts numeric to two decimals and dates ISO, reconciliation flags rows where line items + tax do not match the total within tolerance while clean rows balance (re-check 3 by hand), categories all in-set, low-confidence reads flagged, and the grand total equals the sum of row totals. Write PASS/FAIL per criterion; on any failure, name the receipts and loop back to extract. Then route — this is the whole point of the loop: - **Every criterion PASSES →** call `advance_task_step({ ref, stepId: "evaluate", next: "finish" })`. - **Any criterion FAILS →** write the specific gaps to notes, then call `advance_task_step({ ref, stepId: "evaluate", next: "extract" })` to loop back. The builder fixes exactly those gaps. Never route to `finish` while any criterion is unmet. The build phase's completion gate already blocked a grossly-incomplete deliverable; your job is the judgment an automated check cannot make (does it actually work, read well, look right). After ~3 unproductive loops, stop and report DONE_WITH_CONCERNS so the user can step in. - Finishdeveloper
All acceptance criteria met. Stamp a short summary and report DONE.
Show working prompt
Every acceptance criterion passed. Write a one-paragraph DONE summary to task notes via `write_task_note`: what was built, the deliverable path(s), and a one-line confirmation that each criterion is met. Then report DONE.
Triggers
Phrases that suggest this craftbook to a crew.
- scan my receipts
- receipts to a spreadsheet
- extract receipt totals
- expense report from receipts
- OCR receipts into a ledger
Source
View this craftbook on GitHub · MIT license