accrual-proposals
Skill8 starsChanged 4 months agoMITView on GitHub
Propose month-end accruals from open POs received but not yet invoiced (GRNI), recurring vendor spend patterns, and preparer-supplied one-time items, cross-checked against the AP register so nothing is double-booked, with every line carrying a stated basis and an auto-reversal flag for reviewer approval. Never books anything - output is a proposed JE batch for a Controller to approve before it posts to the GL. Use whenever the user mentions month-end accruals, accrual proposals, what should we accrue this month, GRNI, goods received not invoiced, a recurring accrual estimate, or the accrual JE batch for close. Also trigger on "propose accruals," "draft the accrual entries," "what are we missing on the accrual side," or when an open PO report, AP register, or vendor spend history is supplied together at month end. Inverts cutoff-and-unrecorded-liabilities: that proves a search for what was missed is complete; this proposes what to book. Runs fully local - no client data leaves the machine.
- Installs packages
In this repository44 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- ar-aging-tie-outSkill
- arr-waterfallSkill
- audit-samplingSkill
- audit-workpaperSkill
- balance-sheet-recon-packSkill
- bank-rec-reviewSkill
- bank-rec-to-glSkill
- bank-statement-to-excelSkill
- board-package-builderSkill
- budget-vs-actualSkill
- cash-applicationSkill
- cash-flow-tieoutSkill
- commission-recalcSkill
- consolidation-rollupSkill
- cutoff-and-unrecorded-liabilitiesSkill
- debt-covenant-checkSkill
- deferred-revenue-waterfallSkill
- depreciation-tie-outSkill
- dunning-draftsSkill
- duplicate-payment-scanSkill
- expense-policy-testingSkill
- filing-diffSkill
- flux-narrativeSkill
- fx-revaluationSkill
- intercompany-reconSkill
- invoice-codingSkill
- journal-entry-anomaly-scanSkill
- k1-extract-summarizeSkill
- lease-schedule-tie-outSkill
- month-end-close-checklistSkill
- payroll-tax-reconciliationSkill
- pbc-assemblerSkill
- prepaid-amortizationSkill
- receipts-to-expenseSkill
- return-yoy-varianceSkill
- revenue-trace-to-sourceSkill
- sales-tax-reconciliationSkill
- tax-return-reviewSkill
- thirteen-week-cash-forecastSkill
- three-way-matchSkill
- trial-balance-integritySkill
- vendor-statement-reconSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 29
No clean workpaper unless:
- Very specific: line 75
Every input file except --open-pos, --period-end, and --out is optional — supply what you have.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Month-End Accrual Proposals
cutoff-and-unrecorded-liabilities proves a search is complete and reports what it found.
This skill runs the other direction: it proposes what should be booked, before a reviewer has
asked for it. Same underlying question — what happened before period end that isn't on the
books yet — but the deliverable is a draft journal entry batch instead of a search memo.
Three sources, in order of reliability:
- Open purchase orders with goods or services received but not yet invoiced (GRNI). The most defensible accrual, because it is backed by a receipt, not an estimate.
- Recurring vendor spend — utilities, rent, subscriptions — accrued at the trailing average of prior periods, but only when there is enough history and only when the actual invoice for the current period isn't already recorded.
- Manual one-time items the preparer knows about — legal fees, bonuses, a settlement — each of which must carry its own stated basis. This skill will not accept an amount with no reason attached.
Nothing here gets posted. Every proposed line is a recommendation with a basis and a reversal flag, built for a Controller to approve or reject line by line.
The gate
No clean workpaper unless:
- Every open-PO line, every recurring vendor/category candidate, and every manual item is classified — proposed, excluded with a reason, or an exception with a reason. Nothing is silently dropped, and nothing is silently proposed either.
- Every proposed line has a non-blank basis and an explicit reversal flag. A non-reversing item requires a stated rationale — "no" is never the default.
- Open-PO candidates are cross-checked against the AP register. If the PO report's "invoiced" figure doesn't match what the AP register actually shows, that PO becomes an exception, not a proposal — a stale PO report is the single most common way this kind of tool double-books an accrual.
- Recurring accruals require a minimum amount of history (default three prior periods, configurable) and are dropped to an exception, never guessed, when there isn't enough.
- The proposed batch balances — total debits equal total credits — before it is handed to anyone.
Inputs
- Open PO report — purchase orders with
received_amountandinvoiced_amountin dollars, plus areceipt_date. The difference is the candidate accrual; a missing receipt date makes the line an exception rather than a guess about when it was received. - AP invoice register — invoices recorded through period end, with
po_noso it can be tied back to the open PO report, andvendor/categoryso a recurring estimate can be skipped when the actual invoice already showed up. - Vendor spend history — prior-period actuals by vendor and category, for the recurring accrual average. More periods produce a more defensible average; too few produce an exception instead of an estimate.
- Prior-period accruals, with a
reversedflag — used to flag anything that should have reversed and didn't, which is a double-counting risk this period. - Manual items, if the preparer has known one-time accruals to add — each needs its own basis. This file is optional; nothing here fabricates a manual item on its own.
- Period end date and, only if the close period isn't a calendar month, a period start override. Materiality, if you want the total measured against it.
Step 1 — Run the proposal
python3 scripts/accrual_proposal.py \
--open-pos open_pos.csv --ap-register ap_register.csv \
--vendor-history vendor_history.csv --prior-accruals prior_accruals.csv \
--manual-items manual_items.csv \
--period-end 2026-08-31 --materiality 25000 \
--client "Ardenway Manufacturing" \
--out "Ardenway - August 2026 Accrual Proposals.xlsx"Every input file except --open-pos, --period-end, and --out is optional — supply what
you have. Fewer sources means fewer proposals and more exceptions, not guessed numbers filling
the gap. --min-history (default 3) sets how many prior periods a recurring accrual needs
before it's proposed rather than flagged; --reversal-date defaults to the first day of the
month after period end.
Step 2 — Get the date and the cross-check right
The GRNI accrual depends on when the receipt happened, not the invoice date — a PO receipted 28 August with no invoice yet is an August liability regardless of when the invoice eventually arrives.
The AP register cross-check exists because a PO report's own "invoiced" column goes stale quickly — an invoice keyed against the PO after the report was pulled makes the PO report say "not invoiced" for something that already is. Supplying the AP register lets the script catch that; without it, GRNI candidates are proposed on the PO report's word alone, which the tool says explicitly in its output.
Step 3 — Work the exceptions before the batch goes to the Controller
The Exceptions tab is not noise — it's the population the tool refused to guess on. Common reasons: no receipt date, invoiced-exceeds-received (usually a data error in the PO system), a PO report that doesn't tie to the AP register, insufficient vendor history for a recurring estimate, a manual item with no basis, and a non-reversing manual item with no rationale. Each is worth five minutes before the batch is finalized — most convert into either a proposal or a documented reason to skip it, not a lingering unknown.
Step 4 — Deliver
Workbook tabs:
- Summary — proposal totals by source, materiality comparison, exception count and value, prior-accrual-not-reversed count, the gate results, and sign-off lines for preparer and Controller.
- Proposed Accrual JE Batch — the deliverable: debit and credit accounts and amounts, basis, reversing flag, and reversal date, one pair of lines per proposed entry.
- Auto-Reversal Schedule — the mirrored entries as they'll look on the reversal date, so the reviewer can see the whole lifecycle of each accrual, not just the booking.
- Open PO Detail (GRNI) — every PO line with its classification and notes.
- Recurring Vendor Patterns — the trailing-average calculation, periods used, and classification for each vendor/category.
- Manual Items — the preparer's one-time items, each with its basis and classification.
- Exceptions — every item that could not be proposed, consolidated across sources, with the reason.
- Prior Accrual Reversal Check — prior-period accruals not marked reversed, which is a double-counting risk against anything proposed this period.
Then, in chat: the total proposed, how many exceptions and their value, whether the total exceeds materiality, and a plain statement that this is a proposal only — nothing has been posted, and Controller approval is required before it is booked.
What to escalate
- Any PO where the PO report's invoiced amount doesn't match the AP register. This means the PO report can't be trusted for other periods either until it's reconciled.
- No AP register supplied. GRNI candidates are then proposed on the PO report's own word, with no independent check — say so, don't imply the cross-check ran.
- Prior-period accruals not marked reversed. Booking this period's proposal on top of one that never reversed is a straightforward double-count.
- Recurring categories present in the prior year's pattern and absent from this period's vendor history with no explanation — usually means data wasn't pulled, not that the expense stopped.
- A pattern of PO receipts recorded just after period end that would have been material if dated a day earlier — worth a look at whether the receipt date itself is being entered accurately.
- Non-reversing manual items. These bypass the normal check-and-clear the following month, so the rationale needs to actually hold up, not just be present.
- Total proposed accruals against materiality, and total exceptions against materiality — an exception population large enough to matter is itself a finding.
Security posture
Fully local: standard library plus openpyxl. No network calls, no uploads, no telemetry.
Dependencies
pip install openpyxlBuilt by Adopt AI — free to use and modify.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

