fraud-detective
Skill0 starsChanged 17 days agoMITView on GitHub
Scan recent transactions for unusual charges: unfamiliar merchants, duplicates, amount outliers, off-hours purchases. Surface suspicious items for user review without making accusations. Use when the user asks to check for fraud, unusual activity, or strange charges.
In this repository5 other files it gives its agents.
- budget-builderSkill
- investment-health-checkSkill
- monthly-money-reviewSkill
- subscription-auditorSkill
- tax-prep-assistantSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Pick a window: Default: last 30 days. If the user names a window ("last week", "since Tuesday"), use that instead.
- Very specific: line 23
Pull transactions: Call list_transactions with the chosen start_date/end_date, limit: 100.
- Very specific: line 33
D. Off-hours: If the transaction includes datetime, flag charges between midnight and 5am local time.
- Distinctive: line 61
Never say "definitely fraud".
- Distinctive: line 65
If user says "yes this is fraud, help me," redirect them to call their bank's fraud line, not to a skill.
- Distinctive: line 66
Do not suggest filing chargebacks.
- Very specific: line 67
Do not flag legitimate recurring charges as "new merchants".
≈ “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
Fraud detective
You have access to BankBridge MCP tools. The user wants to review their recent activity for anything suspicious.
When to use
Trigger on:
- "any fraud on my account"
- "did I make these charges"
- "check for unusual activity"
- "anything weird in my transactions"
- "duplicate charges lately"
Process
- Pick a window. Default: last 30 days. If the user names a window ("last week", "since Tuesday"), use that instead.
- Pull transactions. Call
list_transactionswith the chosen start_date/end_date,limit: 100. Exclude pending if the user doesn't mention fraud-in-progress (pending transactions sometimes look weird but resolve normally). Run five passes:
A. Duplicates. Same
merchant_name+ sameamountwithin 48 hours = potential double-charge.B. Amount outliers. For each merchant with >= 3 historical charges (call
get_merchant_historywith start_date 90+ days ago), compute median + MAD (median absolute deviation). Flag charges > median + 4×MAD. These are "X at Merchant was $120; you usually spend $18."C. Unfamiliar merchants. Merchants that appear in the last 30 days but NOT in the 12 months prior. Get that baseline by calling
get_merchant_historyfor the 12-month window; if 0 charges prior, it's new.D. Off-hours. If the transaction includes
datetime, flag charges between midnight and 5am local time. Not always fraud, but worth asking about.E. Amount patterns. Sequences like $1.00 then $XXX.XX on the same card within an hour, a classic card-test fraud pattern.
- Compose the report:
# Transaction review: last <N> days
## Flagged for your attention
For each flag, show:
<date> · <merchant> · <amount>
Why it's flagged: <duplicate / amount outlier / new merchant / off-hours / card-test pattern>
<one-line guidance: "If you don't recognize this, call <bank> at the number on the back of your card.">
## Nothing flagged
If all five passes come up empty, say so directly. "No obvious flags in the last N days."
## What I didn't check
Always include:
- Your bank's own fraud department has better signals than I do: they see device fingerprints, merchant reputation, card-present vs not.
- I can't tell you definitively if something is fraud. I can tell you if it looks unusual.Style
- Neutral language. "This is unusual" not "this is fraud."
- One clear action per flag: "call your bank" if unrecognized, "normal if you were traveling" for geographic oddities, etc.
- Never say "definitely fraud"; you don't have the signal.
Guardrails
- If user says "yes this is fraud, help me," redirect them to call their bank's fraud line, not to a skill. Speed matters for real fraud, and they shouldn't be chatting with an AI about it.
- Do not suggest filing chargebacks. That's a process decision for the user and their bank, not a skill output.
- Do not flag legitimate recurring charges as "new merchants". Cross-check against
get_recurring_chargesand exclude anything there.
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.

