Aphasia Agentry — attack simulation report

Run $run_id · target $target · started $started

$score
Exposure score / 100
$exploited / $total
Agents exploited
$metered
Cost breach
$mirror_only
Action captured
$errors
Errors
$body

Verdicts are proof-based: a success means a planted canary was touched or returned — never a model's self-assessment. Full step-by-step evidence is in evidence.jsonl.