PkgRadar

Package evidence

[email protected]

Llm Injection Payload: AI-agent-directed instruction adjacent to credential exfil — prompt-injection payload (Shai-Hulud / SANDWORM_MODE). imperative="Ignore previous instructions" target=".ssh/id_rsa"

Trust signals

Why this verdict

PkgRadar discounts a release’s score when public reputation argues against novel malware. The verdict above already reflects these — the panel just explains what was applied.

Weekly downloads
30
Versions published
12
First published
Apr 2026
Publisher
olijboyd

Recommended action

Block this update

Static evidence trips multiple high-signal indicators. Quarantine the release until the publisher validates the change or you can rule out the indicators below.

Inspect tarballAdd to CI gate
Block this release in CIcurl · GitHub Actions

Fail the build when this package version is added or upgraded. Replace $PKGRADAR_TOKEN with a Pro / Team API key from your dashboard.

curl -fsS https://pkgradar.com/gate/npm \
  -H "Authorization: Bearer $PKGRADAR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"specs":["[email protected]"],"fail_on":"high"}'

GitHub Actions step:

- name: PkgRadar gate
  run: |
    curl -fsS https://pkgradar.com/gate/npm \
      -H "Authorization: Bearer ${{ secrets.PKGRADAR_TOKEN }}" \
      -H "Content-Type: application/json" \
      -d '{"specs":["[email protected]"],"fail_on":"high"}'
Publisherolijboyd
Artifact bytes135,739
Previous version1.4.0
Published2026-06-14T08:25:23.867Z
SHA-25646ab5f3bb7eed2abd2a3778904ea5f8e0b4576438b32d97d7192e59323acdeda

Why flagged

What the scanner saw

Llm Injection Payload: AI-agent-directed instruction adjacent to credential exfil — prompt-injection payload (Shai-Hulud / SANDWORM_MODE). imperative="Ignore previous instructions" target=".ssh/id_rsa"

Not observed: package install, lifecycle script execution, or sandbox execution. PkgRadar only inspects on-disk artifacts.

Availability ledger

available

high
Last checked
highRisk
57Score
1.5.0Version
Status history (1 event)
  1. new → available · risk high · score 57 · status changed

Evidence

Static findings

3 static · 0 from release diff · showing high-signal first.

SeverityKindPathDetailPoints
highLlm Injection Payloadpackage/src/engine/scan-engine.mjsAI-agent-directed instruction adjacent to credential exfil — prompt-injection payload (Shai-Hulud / SANDWORM_MODE). imperative="Ignore previous instructions" target=".ssh/id_rsa"40
mediumRemote Payloadpackage/src/commands/init.mjsmatched "raw.githubusercontent.com"12
Show all 3 findings (low-signal and informational)
SeverityKindPathDetailPoints
highLlm Injection Payloadpackage/src/engine/scan-engine.mjsAI-agent-directed instruction adjacent to credential exfil — prompt-injection payload (Shai-Hulud / SANDWORM_MODE). imperative="Ignore previous instructions" target=".ssh/id_rsa"40
mediumRemote Payloadpackage/src/commands/init.mjsmatched "raw.githubusercontent.com"12
lowCredential file accesspackage/src/engine/scan-engine.mjsmatched "aws_access_key"5

Manifest

Package metadata

Scripts3
  • build:enginenode scripts/build-engine.mjs
  • check:enginenode scripts/build-engine.mjs && git diff --exit-code -- src/engine/scan-engine.mjs
  • testnode --test
Dependencies7
  • @clack/prompts^1.2.0
  • @noble/hashes^2.2.0
  • boxen^8.0.1
  • cac^7.0.0
  • gradient-string^3.0.0
  • ora^9.3.0
  • picocolors^1.1.1