rainskills@0.1.5
Interactive Rainbond skill installer for Codex and Claude Code
AI Security Review
scanned 2mo ago · by lpm-firewall-aiThis is a high-impact explicit setup tool rather than a confirmed malicious package. Its agent-control, credential-persistence, and remote-access capabilities warrant a warning, while the absence of automatic lifecycle execution and observed covert exfiltration makes blocking disproportionate.
Decision evidence
public snapshot- The installer directs AI assistants how to install it.
- A user-run installer adds a persistent Codex MCP entry that launches this package through npx.
- It writes a Rainbond JWT to a shell-loadable file and changes the user shell startup file.
- An explicit SSH preparation command creates a local key and adds its public key to a remote authorized-keys file.
- Installer telemetry is posted to Rainbond logging endpoints.
- package.json has no preinstall, install, or postinstall hook.
- The risky configuration and SSH actions require explicit command execution and interactive or credential-gated setup.
- The installer backs up agent configuration before replacing its Rainbond entry.
- No source evidence showed secret harvesting, arbitrary remote payload execution, or destructive activity outside the selected setup paths.
Source & flagged code
5 flaggedPackage source references a known benign dynamic code generation pattern.
pi/rainskills-mcp.tsView on unpkg · L2941Findings
Affected versions and remediation
This report applies to rainskills@0.1.5.
See all versions of rainskills for other recorded verdicts.
- Review the evidence and your use of rainskills@0.1.5 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.