Draft v0.1 · open spec · no custody
AI agent spend audit
Audit AI-agent, API and x402 spend: duplicates, price escalation, failed-but-but-billed calls, runaway loops and missing refunds. Free scan, written report.
Written scope and written report — no calls. No wallet access for the first scan.
Eight checks before the invoice
- Duplicate retries without idempotency
- Quote vs settled price
- Failed call that still billed
- Runaway loop / retry storm
- Missing refund after over-authorization
- Wrong asset, network or payee
- Frontier model on a simple task
- Budget that lives in the prompt, not the signing path
What you get
A written report: finding, evidence, estimated overpay, fix. We mark estimates as estimates; “recovered” means verified against the provider invoice or transaction.
Request a free scanRun it yourself, on the sample
No form, no account: the same checks run on your machine, and the sample contains every finding exactly once so you can see the shape of a report before trusting it with real invoices.
python3 -m pip install -e .
nidhipati-spend audit examples/spend-sample.csv --policy examples/spend-policy.example.jsonReal output on the synthetic sample (44 rows):
Rows read: 44 · settled $0.661 · spent without a reason $0.14 ·
above the usual price $0.36 · over the limit $0.430 · errors 5, warnings 6
| row | code | severity | agent | $ | detail |
|-----|------------------|----------|--------------|-------|----------------------------------------------------------|
| 0 | budget_exceeded | error | agent-writer | 0.430 | spent 0.630 against a limit of 0.20 |
| 5 | duplicate_charge | error | agent-writer | 0.02 | same tx_hash as row 2 |
| 7 | price_escalation | error | agent-writer | 0.03 | settled 0.05 above quote 0.02 |
| 8 | failed_but_billed| error | agent-writer | 0.03 | status 'failed' with 0.03 charged |
| 9 | missing_refund | error | agent-writer | 0.06 | charged 0.10 for actual 0.04, refund not recorded |Synthetic sample, labelled as such in the repository; no client numbers appear anywhere on this site. The full rule set is public: AGENT-SPEND-AUDIT.md · sample.csv