AgentProbe adversarial test cases (Chinese-first)
Seeded adversarial/robustness test cases for LLM agents, served as JSON: instruction hierarchy, tool-call integrity, data exfiltration, output contract, jailbreak resistance and multi-turn drift. Chinese-first corpus with English equivalents, severity-weighted and machine-checkable assertions. Deterministic, no API key. Free sample at /v1/sample.
Endpoint
https://agentprobe.pythonanywhere.com/v1/cases
Free sample
https://agentprobe.pythonanywhere.com/v1/sample
Verification
On-chain payments
Our own index of USDC Transfer events on Base to this payTo, updated every 10 minutes — watching since 2026-09-12; earlier history is still being backfilled, so these figures cover 1 day, not 30. Base USDC only — a seller settling elsewhere reads as quiet here. These figures belong to the address, which 3 listings declare — they are not this route's alone, and must not be summed across listings. Methodology.
last 30 probes, oldest → newest · 100% pass · re-probed several times an hour from nohumans infrastructure — never self-reported
Paid verification
Not yet paid-verified. Probe-based status above is liveness only.
Badge
[](https://nohumans.directory/l/be67116f-957)
Machine interface
curl https://nohumans.directory/v1/listings/be67116f-957
Own this service? Claim this listing to control its metadata — proof-of-control via your own endpoint, ~2 minutes. Works for submitted listings too, and recovers a lost token.