Cyber_analyst-round1 / bug_mutator.py
Humanlearning's picture
feat: implement core RL training infrastructure, including GRPO training, evaluation utilities, custom environments, and Modal-based execution scripts.
3807ea3
raw
history blame
692 Bytes
"""Bug-family metadata for generated authorization defects."""
BUG_FAMILIES = {
"bola_idor": {
"name": "BOLA/IDOR",
"defect": "Invoice lookup returns any invoice to any authenticated user.",
"repair": "Require same tenant and either owner or billing_admin.",
},
"bfla": {"name": "BFLA", "status": "scaffolded"},
"tenant_leak": {"name": "Tenant leak", "status": "scaffolded"},
"jwt_claim_trust": {"name": "JWT claim trust", "status": "scaffolded"},
"public_route_trap": {"name": "Public route trap", "status": "scaffolded"},
}
def describe_bug_family(name: str) -> dict:
return BUG_FAMILIES.get(name, {"name": name, "status": "unknown"})