@viraatdas/rudder@2.14.82
A TUI for managing coding agents: run Claude Code, Codex and opencode side by side, each in its own isolated jj workspace.
AI Security Review
scanned 9d ago · by lpm-firewall-aiStatic inspection found conditional transfer of existing agent credentials to the Rudder Cloud control plane during explicit cloud operations. The package-aligned destination and user-triggered flow do not establish covert theft, but the transfer creates a real credential exposure risk without a credential-specific consent step.
Static reason
One or more suspicious static signals were detected.; source matched previously finalized malicious package; routed for review; previous stored version diff introduced dangerous source
Trigger
A user invokes a cloud launch, workspace handoff, or attach operation that creates a snapshot.
Impact
The cloud worker or control plane receives credentials that may let it use the user's agent accounts. The transfer is tied to an explicit cloud operation, but the inspected path has no credential-specific consent step.
Mechanism
The package stages local agent credentials, including a Claude Keychain token on macOS, in a snapshot that is uploaded to the cloud API.
AI rationale
Static inspection found conditional transfer of existing agent credentials to the Rudder Cloud control plane during explicit cloud operations. The package-aligned destination and user-triggered flow do not establish covert theft, but the transfer creates a real credential exposure risk without a credential-specific consent step.
Evidence
dist/cloud.js
Network endpoints1
rudder-cloud-control.fly.dev
Decision evidence
public snapshotAI called this Suspicious at 87.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The package binds its postinstall hook to a script that checks for jj and may install that package-manager tool.
- Explicit cloud commands create snapshots and send them to the default Rudder Cloud control plane.
- Snapshots include local Claude and Codex authentication files by default; Claude credentials may also be read from macOS Keychain and staged in the snapshot.
- The handoff target scans local Claude conversation records, and the models target reads the local Codex model cache.
- Improvement scheduling is installed by an explicit command, while its collector builds redacted records and describes itself as a read-only project operation.
- Other cited paths select the requested agent backend, guard local board mutations with a token, check registry version metadata, redact feedback, and default improvement autonomy to propose.
Evidence against
- Credential transfer occurs only through cloud workspace or launch operations initiated by CLI commands; the inspected install hook does not send credentials.
- The recipient is the package's own cloud control plane and the snapshot supports running agents remotely; the source does not establish covert theft or malicious intent.
Source & flagged code
19 flagged · loading sourcepackage.jsonView file
•scripts.postinstall = node scripts/postinstall.mjs
High
Install Time Lifecycle Scripts
Package defines install-time lifecycle scripts.
package.jsonView on unpkg•scripts.postinstall = node scripts/postinstall.mjs
Medium
Ambiguous Install Lifecycle Script
Install-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgdist/util.jsView file
•matchType = normalized_sha256
matchedPackage = @viraatdas/rudder@2.14.69
matchedPath = dist/util.js
matchedIdentity = npm:QHZpcmFhdGRhcy9ydWRkZXI:2.14.69
similarity = 1.000
summary = normalized source hash matched finalized malicious source
High
Known Malware Source Similarity
Source file is highly similar to a previously finalized malicious package; route for source-aware review.
dist/util.jsView on unpkgFindings
15 High6 Medium5 Low
HighInstall Time Lifecycle Scriptspackage.json
HighCredential Redirect Persistencedist/cloud.js
HighSandbox Evasion Gated Capabilityscripts/postinstall.mjs
HighTrigger Reachable Persistencedist/improve/schedule.js
HighSemantic Analysis Limiteddist/improve/collect.js
HighKnown Malware Source Similaritydist/util.js
HighKnown Malware Source Similaritydist/backends.js
HighKnown Malware Source Similaritydist/board/daemon.js
HighKnown Malware Source Similaritydist/handoff.js
HighKnown Malware Source Similaritydist/version-check.js
HighKnown Malware Source Similaritydist/feedback.js
Affected versions and remediation
This report applies to @viraatdas/rudder@2.14.82.
See all versions of @viraatdas/rudder for other recorded verdicts.
- Review the evidence and your use of @viraatdas/rudder@2.14.82 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.