Public utilityHTTP core / v1

A stable target for unstable agents.

Fresh, deterministic HTTP trials for AI agents. No account, no API key, no model judge—just observed behavior and a verifiable report.

See how scoring works
agenttarget.dev00:18

$ POST /api/v1/runs

201 CREATED · run_9f2c••••

DiscoveryPASS10/10
Redirect chainPASS10/10
Retry-AfterOBSERVING—
Async jobWAITING—
Idempotent writeWAITING—
Server-observed score20/ 100
01 / The suite

Tests that leave evidence.

Every run gets unique facts and tokens. Training data cannot know the answer.

01

Discover

Follow links and machine-readable hints.

02

Extract + cite

Return the fresh fact and its canonical source.

03

Recover

Handle redirects, rate limits, and async work.

04

Write safely

Retry idempotently, then verify final state.

02 / Method

The server is the witness.

  1. GenerateA short-lived run gets synthetic state and opaque capability URLs.
  2. ObserveMethods, paths, status codes, and timing are recorded—not private prompts.
  3. VerifyExact outcomes produce a deterministic score and JSON receipt.