@minimax-ai/code@0.4.12
⚠ Under reviewMinimax Code — terminal coding agent
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThe only install hook is a local better-sqlite3 compatibility check, and the flagged agent sites read or label files during normal CLI use. A clean publication is blocked because the required citation list is longer than the citation cap, and no independent malicious chain was found.
Decision evidence
public snapshot- package.json defines only a postinstall that runs verify-native-install.mjs. Dependencies are clipboard, ripgrep, and optional better-sqlite3, not this package name.
- That postinstall checks the Node and glibc range, then smoke-tests optional better-sqlite3 and prints reinstall text. It does not write agent config or open a network connection.
- Launcher code reads recent local Codex session files under CODEX_HOME or ~/.codex and returns a session id when the workspace matches. It is not an npm lifecycle hook.
- Tool-output code treats AGENTS.md and CLAUDE.md as labeled read targets. That is display classification, not a config write.
- A clean decision would need citations for 10 required paths, but sourceCitations allows 8. No separate install-time or import-time attack chain was found in the inspected entrypoints.
- There is no preinstall or install script and no runtime dependency on @minimax-ai/code.
- The PDF skill uses child_process only to locate or install Playwright Chromium when that skill script runs.
Source & flagged code
25 flaggedPackage defines install-time lifecycle scripts.
package.jsonView on unpkgPackage source references child process execution.
chunks/chunk-2CEKQTMO.jsView on unpkg · L1Findings
Affected versions and remediation
This report applies to @minimax-ai/code@0.4.12.
See all versions of @minimax-ai/code for other recorded verdicts.
- Review the evidence and your use of @minimax-ai/code@0.4.12 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.