The Manifest

CHAINDROP and the full installed universe: catching a 400-package worm

CHAINDROP trojanized 400-500 npm packages through one compromised maintainer. We mapped the campaign family against our engines, found two blind spots in our own malware shield, fixed them, and verified the coverage end to end.

P

Pinhaul Crew

October 9, 2026 · 3 min read

August 2026: a worm descendant security researchers named CHAINDROP trojanized 400–500 npm packages by compromising a single maintainer of two of the most widely depended-upon cache libraries in the ecosystem. It was the fourth supply-chain campaign in as many months — and it gave us the perfect test case for a claim we make on our landing page: any scanner can subscribe to the feeds; Pinhaul cross-references them against your live inventory and moves before they do. This post is the receipts.

The campaign, compressed

The family tree: Shai-Hulud (September 2025, the first self-replicating npm worm, [CISA-alerted](https://www.cisa.gov/news-events/alerts/2025/09/23/widespread-supply-chain-compromise-impacting-npm-ecosystem)) → Mini Shai-Hulud (2026: 84 @tanstack artifacts in May, @antv packages in May–June, 32 more packages in June) → CHAINDROP (August: the keyv/cacheable maintainer compromised, 400–500 packages). Alongside them: typosquatted payment SDKs sweeping environment variables for tokens, and a stolen publisher token backdooring jscrambler — executing on import, so npm install --ignore-scripts offered no protection.

Each campaign has OpenSSF MAL records in the npm OSV export. We checked: 331 MAL records across the campaign package families, including MAL-2026-11524 (keyv) and MAL-2026-11963 (cacheable). The data was there. The question is what a scanner does with it.

The gap we found in our own shield

Honesty first: when we mapped these campaigns against Pinhaul, the malware engine had two blind spots.

Agent-only hosts had no coverage at all. The malware engine matches the feed against your installed-package inventory — but inventory rows come from the SDK. A host watched only by the agent had Trivy findings but an empty inventory, so there was nothing to match against. Your agent-covered hosts — the ones most likely to be running production workloads — were the least covered.

Transitive dependencies were filtered out. The shield only matched direct pins. Worms do not respect that: CHAINDROP and friends compromise packages you depend on indirectly, exactly like the jscrambler case — a build-time dependency pulled in by your real dependencies.

The fix

The malware engine now matches against the full installed universe: runtime inventory, direct and transitive, plus every package/version our container scans have ever observed on the host. A known-malicious package anywhere in the graph — any depth, any container, discovered by any engine — becomes a critical finding with the MAL identifier attached. A package in the feed but not installed never bothers you.

We verified the pipeline end to end on production: ingesting the campaign as intel records, then adding a jscrambler 8.21.0 dependency (one of the poisoned versions, fixed in 8.22.0) to a test project. The rematch produced a critical finding immediately, linked to the source article.

The speed play

Feed lag is a security control nobody talks about. Between disclosure and feed ingestion there is a window — hours to days — where every subscription-based scanner is blind. Our intel pipeline closes it: an article like the one describing these campaigns goes through a model-reviewed intake, becomes structured records in minutes, and matches against every customer’s inventory immediately. During the jscrambler window that was the difference between knowing and not knowing.

What we take from this

  • Malware is a position problem. A malicious package two hops deep in your graph is as dangerous as one you installed yourself. Scan the whole graph.
  • Agents and SDKs must agree. If your monitoring only covers hosts with the SDK installed, your servers are the blind spot. One universe, both sources.
  • Provenance is not safety. CHAINDROP shipped with valid Sigstore and SLSA attestations because the attacker owned the CI identity. Signatures prove where code came from — not that it was safe.

All of this is running now. If you want it watching your stack — join the waitlist; we onboard a few builders every week.

Want this watching your stack?

Every engine in this story runs nightly for the teams on Pinhaul. Join the waitlist — we onboard a few builders every week.