Plumbline: A Graph-Powered Agent for Supply-Chain Incident Response

Description and use

Plumbline leverages the FalkorDB graph database to automatically analyze npm lockfile dependencies, issuing deterministic EXPOSED, CLEARED, or UNKNOWN verdicts with evidence and hashed receipts, while tracking fact changes and automatically reopening affected decisions.

Problem

During a software supply‑chain incident, teams need to know which services are affected by a poisoned package and how the malicious version propagates through dependencies, rather than just whether a package is vulnerable.

Outcome

Plumbline provides deterministic, evidence‑backed verdicts, transparent receipts, and decision memory that automatically reopens affected approvals when new evidence arrives, improving incident response accuracy and reproducibility.

Workflow

• 1. Upload an npm lockfile to the lockfile checker, creating an isolated FalkorDB graph.
• 2. Plumbline queries the graph to find repositories that lock affected package versions.
• 3. It computes the shortest dependency path from the service to the malicious package.
• 4. The deterministic verdict policy assigns EXPOSED, CLEARED, or UNKNOWN based on evidence and assumptions.
• 5. A receipt containing the verdict, evidence, path, assumptions, and policy version is generated and hashed.
• 6. Decision memory links approvals to the facts they assume; when a fact changes, only affected decisions reopen.
• 7. The MCP server exposes tools for triage, path explanation, fix ranking, and task queue management.
• 8. Users review receipts, inspect paths, and approve decisions via the UI.

Evidence

Plumbline is an open‑source supply‑chain triage agent. It follows dependency paths in FalkorDB, produces evidence-backed EXPOSED, CLEARED, or UNKNOWN verdicts, remembers the assumptions behind human decisions, and reopens only the decisions affected by new evidence.

Source

Plumbline produces one of three verdicts: EXPOSED, CLEARED, or UNKNOWN.

Source

Each verdict has a receipt containing the verdict, evidence, dependency path, assumptions, and policy version. The receipt is hashed and can be re‑derived from the graph.

Source