chemx@26.10.8-344
Chemical X Protocol: Private drop-in architecture starter kit and capsule generator
AI Security Review
scanned 5d ago · by lpm-firewall-aiThe postinstall hook automatically changes project and agent MCP configuration, leaving a concrete extension lifecycle risk that warrants a warning. The inspected source shows first-party server registration, not a confirmed attack requiring a block.
Decision evidence
public snapshot- The package manifest defines a postinstall hook that runs cli/postinstall.js.
- The postinstall hook automatically calls installAllMcpConfigs for the installation target, except when installing inside the starter kit itself.
- The installer writes an MCP server entry to the consumer project’s .cursor/mcp.json; the required installer location covers the start of this configuration path.
- The Antigravity setup merges and writes a server entry to the user’s global MCP config when that config directory exists.
- The registered server runs this package’s local MCP entrypoint when available, otherwise it invokes chemx mcp through npm exec; Antigravity also receives tool schemas and instructions.
- The inspected behavior establishes automatic first-party agent configuration, but does not establish credential transfer, permission grants, or an unrelated payload.
Source & flagged code
14 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 child process execution.
cli/audit-preflight-git.jsView on unpkg · L1Findings
Affected versions and remediation
This report applies to chemx@26.10.8-344.
See all versions of chemx for other recorded verdicts.
- Review the evidence and your use of chemx@26.10.8-344 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.