Failure Reproduction

Each fuzz campaign starts from a base seed. Individual iterations derive from seed + n.

When a fuzzer fails, as-test reports:

• the base repro command
• the exact failing seeds
• one-run repro commands like ast fuzz ... --seed <seed+n> --runs 1
• the captured run(...) inputs in .as-test/crashes

Why This Matters

This makes failures easier to replay in two different ways:

1. rerun the exact failing iteration by seed
2. inspect the captured input payload even if the generator has side effects

Example Output

Failing seeds: 8, 9, 10
Repro 2: ast fuzz assembly/__fuzz__/parser.fuzz.ts --seed 8 --runs 1
Input 2: [0]

Crash Artifacts

Crash metadata is written under the configured crashDir, which defaults to ./.as-test/crashes.

Each failure record includes:

• mode
• first failing seed
• repro command
• first assertion failure details
• the list of failing seeds and captured inputs