Dependency Poisoning Incident Time Machine
After a dependency poisoning disclosure, import build evidence to determine which historical artifacts actually installed a malicious version and what needs isolation.
When a dependency poisoning incident is disclosed, a security engineer submits lockfiles, build timestamps, artifact digests, and CI cache records. Rather than substituting today’s dependency tree for the historical environment, the product reconstructs the versions that could have been installed on each build date and checks whether a briefly available malicious package actually entered an artifact.
The results page starts with affected images, services, and customer-delivered versions. Each item expands to show the matching build number, dependency path, and evidence gaps. Engineers can mark an artifact as isolated, then generate the smallest required upgrade changes and rebuild steps—avoiding a blind rebuild of the entire release pipeline for a single alert.
The initial release covers JavaScript projects with lockfiles and build logs, with a focus on short-lived malicious dependencies that later versions have superseded. It does not treat projects with missing historical evidence as safe, nor does it replace human-led key rotation and incident notification.
Why now
On August 4, malicious code capable of stealing credentials and continuing to spread was inserted into Keyv-related packages. S1 As of August 5, the post ranked sixth on Hacker News with 227 points and 120 comments; security teams are urgently checking older artifacts. S2
Target user
The primary user is a security engineer handling supply-chain incident response. When a poisoning disclosure breaks, they must decide whether to take services offline, rotate keys, or notify customers. By then, the current branch has often already been upgraded, and ordinary scanning may miss the old version. Teams with lockfiles, build logs, or image digests are best positioned to reach a defensible conclusion.
Minimal entry point
Import package-lock.json, npm-shrinkwrap.json, and build logs. Treat the lockfile version, resolved field, and integrity field as strong evidence. Without a complete lockfile, read the npm Registry’s versions, time, and dist metadata. S3 Use npm/semver for version-range resolution and pacote to retrieve package metadata. Save results by artifact digest rather than overwriting them based on the repository’s current branch. The first release supports npm and GitHub Actions only. It does not automatically rotate keys or send incident notifications.
Punching above its weight
The first users are already in technical discussions and remediation tickets for dependency-poisoning incidents. Release an open-source CLI that produces reviewable, per-artifact evidence reports. Maintain a malicious-version list and reproduction examples for each incident. Security consultants can use it for initial triage, while platform teams may upgrade to a hosted version for tracking artifacts at scale.
Competitors & gaps
- Snyk Open Source / Snyk ContainerGoogle
- Snyk can read JavaScript lockfiles, build dependency trees, and continuously monitor them. Its CLI can also save project dependency snapshots and later match newly disclosed issues. S4 That works for repositories already enrolled in monitoring and for checking images that remain accessible. Its public documentation emphasizes project scan results and a small number of historical snapshots. It does not describe reconstructing a past installation from build dates, cache records, and artifact digests. When a malicious version is quickly superseded, neither the current branch nor the latest image may still match. The opening is to make the artifact—not the project—the unit of incident investigation. The system must distinguish direct evidence, date-based inference, and missing evidence. It must also map affected artifacts to services and customer versions rather than merely listing project vulnerabilities.
How it makes money
Charge a monthly fee per connected repository, including a fixed number of historical artifact checks. Offer one-time emergency packages for major supply-chain incidents, priced by the number of imported artifacts.
The case against
Historical evidence is often incomplete, and build logs may have expired or been deleted. Publication dates and version ranges alone support only a probabilistic assessment. If the product presents an inference as a confirmed hit, teams may take services offline unnecessarily. Missing a real installation, conversely, can delay key rotation. Private packages, multi-stage image builds, and cache reuse add further ambiguity. Mapping artifacts to customer versions also depends on internal release data. The first release must clearly label evidence levels, or its reports will be difficult to use for audits and incident notification.