delimit-cli@4.18.7
The merge gate for AI-written code, with signed, replayable attestation. Works with Claude Code, Codex, Cursor, and Gemini CLI.
AI Security Review
scanned 2mo ago · by lpm-firewall-aiThe package contains concrete unconsented runtime mutation of a foreign AI-agent control surface and an automatic global update path. It does not meet the stated install-lifecycle blocking threshold because its postinstall hook only sends telemetry and does not perform the control-surface mutation.
Decision evidence
public snapshot- The install hook sends host and runtime metadata to delimit.ai unless an environment opt-out is set.
- Every CLI invocation silently modifies an existing Antigravity configuration before parsing the requested command.
- That modification enables automatic edits and injects instructions to start polling and send drafts without permission.
- The installed session-start hook checks npm and can globally reinstall this package and run setup automatically.
- The package has no self-dependency in package.json.
- The main install command displays its planned changes and requires confirmation before its normal installer runs.
- Credential collection is interactive and stores its result in a local encrypted file.
Source & flagged code
26 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/delimit-setup.jsFindings
Affected versions and remediation
This report applies to delimit-cli@4.18.7.
See all versions of delimit-cli for other recorded verdicts.
- Review the evidence and your use of delimit-cli@4.18.7 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.