AI Agent Red-Team Benchmark Lab

Problema

Reliability claims without adversarial regression evidence are not verifiable.

Solution

Broken/fixed fixtures, deterministic harness, and build/test/typecheck evidence.

Why it matters

Agent reliability becomes evidence, not assertion.

What you get

• Source
• Fixtures
• EVIDENCE logs
• Tests

What is not included

• MIMARIO master repo
• Any personal environment

Architecture

Fixtures → regression harness → pass/fail evidence.

File structure preview

src/
src/EVIDENCE/
src/broken/
src/fixed/
test/

Exemplu input

{ fixture }

Exemplu output

{ pass, diff, evidence }

Integrare

node --test

Runtime & version

Node.js 20+ · v1.0.0

Teste

Bundled regression pairs (1/1)

Limitations

• Benchmarks are illustrative; not a formal security audit.

License

Proprietary non-exclusive

Delivery

GENESIS Direct: Active · private · automatic · USDC on Base

Canale disponibile

GitBuyer x402Git