@getstackrun/sdk@1.2.0
Official JavaScript/TypeScript SDK for STACK - the runtime control plane for AI agents
AI Security Review
scanned 3d ago · by lpm-firewall-aiThe flagged subprocess behavior is limited to opening the browser during authentication or agent enrollment. Credential refresh and enrollment requests go to the configured STACK API, and the inspected source does not execute commands or forward command output.
Decision evidence
public snapshot- The package exports its SDK entrypoint from dist/index.js; the only lifecycle script is prepublishOnly, which runs clean and build.
- The Stack constructor creates an HTTP client and service objects; this is the SDK’s expected entrypoint behavior.
- Refresh-token exchange sends the stored STACK refresh token to the configured API base URL’s /oauth/token endpoint.
- Agent enrollment sends a challenge response and public key to the configured STACK API; the private key is stored locally.
- The interactive enrollment flow opens the STACK dashboard and includes the agent ID, machine name, and process label in the enrollment URL.
- The package uses child_process, but the inspected calls only open a browser for authentication or enrollment.
- The SDK reads local STACK credentials, but the inspected refresh flow sends them to the configured STACK API.
Source & flagged code
5 flaggedThis package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
dist/browser-bootstrap.jsView on unpkgPackage source references child process execution.
dist/browser-bootstrap.jsView on unpkg · L181Findings
Affected versions and remediation
This report applies to @getstackrun/sdk@1.2.0.
See all versions of @getstackrun/sdk for other recorded verdicts.
- Review the evidence and your use of @getstackrun/sdk@1.2.0 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.