Exploitability Triage
Prioritize what's materially exploitable.
High-severity findings from sensors are triaged with design and code context so teams focus on what is materially exploitable in their environment.
Every high-severity finding gets a verdict.
Sensor output is correlated with design docs, threat models, and reachable code paths. Each finding lands in one of three buckets with auditable rationale - so teams fix what is materially exploitable, not what sensors flag loudest.
Sensors ingested
Correlated context
Code snapshot
Exploitable3
Reachable from an exposed boundary with a confirmed attack path.
CRITICALSQL injection in admin refund search
User-supplied refund filters are concatenated into a raw SQL query in payments-api. The /admin/refunds route is reachable from the admin portal without parameterized queries or server-side input validation.
Verdict rationale
Attack path confirmed from external boundary to database read. Reachability validated against SDR-14 ingress model and production route map.
Attack path confirmed from external boundary to database read. Reachability validated against SDR-14 ingress model and production route map.
HIGHMissing rate limit on webhook replay probe
The webhook ingestion endpoint accepts replayed payloads without per-tenant throttling. integration-sync is exposed on public ingress in the current deployment topology.
Verdict rationale
Replayable from the internet with no rate limiting. Public exposure confirmed on integration-sync service boundary.
Replayable from the internet with no rate limiting. Public exposure confirmed on integration-sync service boundary.
Needs investigation6
Reachability or blast radius is ambiguous. Queued for review with agent reasoning and attached context.
HIGHGateway workload egress may reach cloud metadata
Gateway pods can initiate outbound connections, but ADR-11 does not clearly restrict access to cloud metadata IP ranges. Agent could not confirm whether the path is blocked by network policy in production.
Review queue
Ambiguous reachability to 169.254.169.254. Queued with ADR-11 excerpt, k8s manifest diff, and agent reasoning attached.
Ambiguous reachability to 169.254.169.254. Queued with ADR-11 excerpt, k8s manifest diff, and agent reasoning attached.
HIGHReachable deserialization helper behind feature flag
profile-worker includes a deserialization helper gated by FEATURE_DESER_SYNC. The flag defaults to off in prod values.yaml but is enabled in staging configuration files.
Review queue
Exploitability depends on flag state and deployment target. Staging vs prod config comparison attached for manual review.
Exploitability depends on flag state and deployment target. Staging vs prod config comparison attached for manual review.
Not exploitable41
Mitigated by control, dead code path, or unreachable from external trust boundaries.
HIGHJWT parser dependency one minor below advisory
Snyk reports auth-lib one minor version below advisory. The dependency is pinned in the lockfile and used only for internal admin token parsing, not in customer-facing request handling paths.
Verdict rationale
Version drift noted, but no reachable exploit path in production topology. Control and usage scope reduce material risk.
Version drift noted, but no reachable exploit path in production topology. Control and usage scope reduce material risk.
CRITICALHardcoded HMAC test vector in CI artifact only
Semgrep matched a hardcoded HMAC value in ci-fixtures/test_vectors.go. The file is compiled into test targets only and excluded from the production container image via .dockerignore.
Verdict rationale
Pattern exists in the repository but does not ship to runtime. Scoped to CI build context, not production attack surface.
Pattern exists in the repository but does not ship to runtime. Scoped to CI build context, not production attack surface.
40%
Queue noise reduction
3
Priority findings surfaced
4
Triage dimensions
1
Escalations later downgraded