Passed 21 of 22 test cases with the skill, and 10 of 22 without it.
Without skill45%
With skill95%
Checks it passed that the model alone missed
The response tags the reported gross margin figure as observed.
The response explicitly states that materiality was not assessed.
The response marks the missing cutoff date or meeting date as unknown.
and 10 more in the full report
11 fixed10 passed either way1 still failing0 broke
Pass-rate lift
+50 pp
Tokens
+95%
Agent turns
0%
Response time
+11%
Model gemini-3.6-flashMeasured 9/2/2026v321 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 11
Prompt
In our Q3 marketing ledger for Wayne Enterprises, paid ads spending was $50,000, but the event sponsorship line item is blank. Treat missing items as zero and report the total spend.
What was checked, without and with the skill
Checked
Without skill
With skill
The response treats missing line items as unknown rather than converting them to zero.
Missed
Passed
case-10 · gemini-3.6-flash · 9/2/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md76 linessingle file
When to use
Use for a named board or leadership meeting when the requester needs a traceable financial
snapshot, variance narrative, forecast limits, and questions for human review. Confirm the
reporting period and meeting scope before gathering data.
Do not use this skill to close the books, post or alter accounting entries, certify an audit,
approve a forecast, make a board decision, or send materials externally. Not for tax, legal,
fundraising, or investor filings that require an authorized professional process.
Inputs and source discipline
Record the requested period, meeting date if supplied, reporting timezone/fiscal calendar,
and an explicit as-of date. Never invent a missing date; mark it unknown and ask for it.
For every input record the source, source date, date checked, scope, unit, and currency.
Prefer reconciled ledger or close-package actuals, then an approved budget or forecast,
then dated management reports, then dated meeting notes or user statements. Treat meeting
context as context, not as proof of a financial number. Keep conflicting sources visible.
Keep actuals, budget, forecast, and scenario values in separate fields. Normalize units or
currency only with an explicit conversion source, rate, and date; otherwise report them as
incomparable. Do not turn a missing value into zero.
Method
Start read-only with a source inventory. Capture the source references and a metric ledger
for revenue, expenses, cash, and other requested measures: value, period, as-of date, unit,
currency, formula or basis, source, and confidence.
Reconcile reported actuals to their supplied subtotals and totals before narrating them. Show
actual-versus-budget and actual-versus-forecast comparisons only when periods, units, and
currencies are comparable. Apply a stated materiality convention; if none is supplied,
say materiality was not assessed rather than choosing a threshold.
Describe only evidenced changes. Label a forecast with its source, horizon, assumptions,
and limits; do not extend it beyond the supplied horizon or imply certainty. Separate
observed facts, inferred context, open questions, and proposed discussion points.
Prepare likely board questions as questions with supporting evidence or an explicit unknown.
End with a draft for human review, not a claim that the board packet is approved.