@sciagent/cli@1.0.83
SciAgent CLI - AI Research Assistant
AI Security Review
scanned 3mo ago · by lpm-firewall-aiThe package performs unconsented install-time global package installation and unverified executable staging, creating material remote-code risk. Source inspection does not prove an active malicious objective, so a warning is more proportionate than a block under the stated policy.
Decision evidence
public snapshot- `package.json` runs `node scripts/postinstall.js` automatically.
- `scripts/postinstall.js` downloads native CLI and CodeBuddy binaries during postinstall without signature/hash verification.
- Postinstall executes `npm install -g @tencent-ai/codebuddy-code` and writes executable wrappers under the user profile.
- `bin/sciagent.js` re-downloads an executable via curl/PowerShell when the CLI runs, then executes it.
- Downloaded payloads are accepted by size only and come from mutable remote endpoints.
- No source reads environment secrets, SSH files, browser data, or credentials for exfiltration.
- No observed AI-agent configuration/control-surface mutation outside the package-owned `.sciagent` directory.
- No persistence beyond package-owned binaries, wrappers, and version marker.
- No obfuscation, eval/vm, or unrelated destructive file operations found.
Source & flagged code
6 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgThis package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
bin/sciagent.jsFindings
Affected versions and remediation
This report applies to @sciagent/cli@1.0.83.
See all versions of @sciagent/cli for other recorded verdicts.
- Review the evidence and your use of @sciagent/cli@1.0.83 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.