sealclaw@2026.9.23
Multi-channel AI gateway with extensible messaging integrations
AI Security Review
scanned 15d ago · by lpm-firewall-aiThe inspected install hook only warns about the detected package manager, and the flagged CLI excerpt creates wiki documentation files. The right-to-left override in the control UI bundle remains unexplained, so a warning is appropriate without claiming an active attack.
Decision evidence
public snapshot- The control UI bundle contains a right-to-left override character amid unusual Unicode data; static inspection did not establish its purpose.
- The preinstall hook detects the package manager and prints a warning for non-pnpm installs.
- The flagged CLI code writes generated AGENTS.md and WIKI.md files; this excerpt does not show agent configuration or plugin installation.
- The audit code inspects installed plugin directories, while the plugin SDK contract detects import-time side effects.
- The flagged Python file is a test helper that loads scripts for tests.
- The manifest’s preinstall command runs the package-manager warning script; the inspected warning code does not mutate files or contact a network endpoint.
- No concrete credential collection, exfiltration, persistence, or destructive behavior was established in the inspected excerpts.
Source & flagged code
12 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 unpkgPackage source references dynamic require/import behavior.
dist/channel-entry-contract-D5iL7RmB.jsFindings
Affected versions and remediation
This report applies to sealclaw@2026.9.23.
See all versions of sealclaw for other recorded verdicts.
- Review the evidence and your use of sealclaw@2026.9.23 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.