Design a budget-vs-actuals tracker that stays alive past February — the category grain that matches real statements, the variance view that answers 'are we okay', the update ritual small enough to survive, and the honest handling of irregular expenses.
SkillScore82Maintained
Automatic
◆Finance & Accounting pack
Last verified 8/14/2026
What it changes
Passed 18 of 22 test cases with the skill, and 6 of 22 without it.
Without skill27%
With skill82%
Checks it passed that the model alone missed
The process finishes with writing a single summary sentence.
The system incorporates both month and YTD variance columns.
The framework restricts total main categories to between 8 and 15.
and 18 more in the full report
12 fixed6 passed either way4 still failing0 broke
Pass-rate lift
+55 pp
Tokens
+45%
Agent turns
0%
Response time
+25%
Model gemini-3.6-flashMeasured 8/14/2026v118 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 12
Prompt
I'm tired of our family budget setup falling apart every few months. We spend money via two credit cards and a primary checking account, and we constantly get caught off guard by annual property taxes and semi-annual car insurance. Can you design a functional budget tracker blueprint for us? I need to see a category and monthly grid layout, a framework for comparing actual spending against budget with variance warnings, a table to manage non-monthly recurring expenses, and a simple monthly review process.
What was checked, without and with the skill
Checked
Without skill
With skill
The response includes both current month and year-to-date (YTD) variance comparison columns.
Missed
Passed
The non-monthly recurring expenses table includes columns for lump item, billing timing, annual total, and monthly accrual amount.
Passed
Passed
The monthly review routine specifies concluding with exactly one summary sentence.
Missed
Passed
case-01 · gemini-3.6-flash · 8/14/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md59 linessingle file
Budget Tracker Design Skill
Budget trackers die two deaths: too granular (forty categories nobody can code a coffee receipt into — abandoned by February) or too vague (three buckets that hide every overrun until it's structural). The living tracker matches its grain to how spending actually arrives (statement lines, invoices), answers one question at a glance — "are we okay, and where not?" — handles irregular expenses honestly (the annual insurance bill is not a "shock"), and costs fifteen minutes a month, because the ritual's survivability is the design's real constraint.
What This Skill Produces
The structure — categories at statement-grain (8–15, not 40), the month × category grid, budget column with its source
The variance view — actual vs. budget with the month-and-YTD dual read, conditional signals at honest thresholds
The irregulars ledger — annual/lumpy expenses spread as monthly accruals so they stop being surprises
The ritual — the fifteen-minute monthly update, sourced from statements, calendared
Required Inputs
Ask for these if not provided:
The scope — household, team, or project budget; and the currency of pain ("we get surprised" vs "we can't approve spend" vs "the money just goes")
The spend sources — cards, invoices, payroll, reimbursements; categories must match how these report, or every update becomes archaeology
The budget numbers — from where (last year +X%? A plan? First-time guesses marked as guesses to be re-based at month 3)?
The irregulars — the annual/quarterly lumps (insurance, subscriptions (subscription-audit finds them), taxes, conferences) — listed, because the design accrues them
Framework: The Design Rules
Grain follows the statement: categories are what spending arrives labeled as — if the card statement says "Software," a tracker with nine software sub-categories requires a coder, and the coder quits. 8–15 categories; anything needing more is two budgets.
The dual read is the view: this-month variance (the early warning) AND year-to-date variance (the truth) — monthly noise means YTD is where "are we okay" actually lives, and trackers showing only months train panic-then-complacency cycles.
Accrue the irregulars: every lumpy expense becomes a monthly twelfth in its category (the insurance is $60/month that bills annually) — the single design move that ends the "shock" pattern, because the shocks were scheduled all along. The irregulars ledger lists each lump, its month, and its accrual.
Thresholds mean something: the variance signal fires at a level worth attention (±10% or a real amount, scaled to the category) — a tracker that flags $3 overruns in red teaches signal-blindness by March.
The ritual is the moat: monthly, 15 minutes, from statements: fill actuals (statement-grain makes this transcription, not judgment), scan the dual read, write ONE sentence ("okay except software, +22% YTD — the new tool stack"). The sentence is the tracker's product; a tracker updated but unread is a diary, and one demanding an hour is abandoned.
Output Format
Budget Tracker: [scope] — [year]
Structure
[Categories (statement-grained) × months · budget column + source note · the guesses marked for month-3 rebase]
Variance View
[Month + YTD columns · signal thresholds per category · the one-glance "are we okay" row]