Vulnerability Chaining & Choke Points

🔐 Enterprise Tier · Cross-rule attack-path composition

Individual findings rarely tell the whole story. SquireX's chaining engine (ComposeChains) composes multiple graph-anchored findings into a single attack path, escalates its severity, and names the one node that — if remediated — breaks the most chains: the choke point.

This is the difference between "here are 12 medium findings" and "these 3 findings form one critical, reachable path — fix this one node to collapse it."

---

How a chain is composed

A set of findings is promoted to a chain only when all gates pass — chains are never speculative:

1. Graph-anchored — every constituent finding is anchored to a Semantic Graph node (Violation.nodeId).
2. Reachable — a path exists from an attacker-controllable agent entry point to a sink (reachability-gated BFS).
3. Crosses a trust boundary — the path traverses ≥1 trust-boundary edge (e.g. agent → tool, agent → external callout). Findings that never cross a boundary are not chained.

If any gate fails (unreachable, or no trust-boundary crossing), the findings remain standalone — no false attack path.

Severity escalation

A reachable, boundary-crossing chain is more dangerous than its parts. Severity escalates one level per additional constituent, capped at critical:

Constituent findings	Composed severity
3 × medium on one reachable path	critical
2 × medium	high
2 × low	medium

Escalation is deterministic — the same graph always produces the same chain and severity.

The choke point

For every chain, the engine computes a global choke-point frequency: the anchored node that appears across the most chains. Remediating the choke point collapses the largest number of attack paths for the least work — SquireX surfaces it explicitly so you fix the root, not the symptoms.

graph LR
  ENTRY["Agent entry\n(attacker-reachable)"]
  A["Finding A\n(medium)"]
  CHOKE["⛔ Choke point\n(shared node)"]
  B["Finding B\n(medium)"]
  SINK["Sink\n(external callout / DML)"]

  ENTRY --> A --> CHOKE
  CHOKE --> B --> SINK
  CHOKE -.->|"breaks the most chains if fixed"| CHOKE

---

In the SARIF output

Each constituent finding is kept as its own result (nothing is hidden), and a synthesized chain result is added with:

• relatedLocations — every constituent finding's location, so the path is navigable in GitHub Advanced Security and the VS Code viewer
• properties.chain — the ordered node path
• properties.chokePoint — the node to remediate first
• properties.constituentRuleIds — the rule IDs that composed the chain

The CI-Native Agent reads chokePoint to target the highest-leverage fix first. See SARIF Output.

---

Why it matters

This is the paid differentiator: SquireX doesn't just list issues, it reasons about how they combine into a reachable, trust-boundary-crossing exploit — and tells you the single highest-leverage place to break it.