Codex CLI plugin that turns a request into a goal with checks, keeps the evidence in your project and hands unfinished work to the next session.
Allowed — no malicious behavior detected
19 low-signal patterns reviewed and cleared.
AI security review
scanned · by lpm-firewall-ai
Source inspection resolved the scanner hints: the lifecycle behavior is a gated, package-aligned cache prewarm, and plugin or agent setup is routed through the explicit CLI install flow. The flagged document, test, and page-probe files do not establish malicious behavior.
A global npm install outside CI triggers the postinstall prewarm.
Impact
No confirmed malicious effect was found.
Mechanism
The prewarm stages pinned runtime dependencies and assets in the product cache; no credential transfer or automatic agent configuration write was identified.
Decision evidence
public snapshot
Evidence for AI clean decision
package.json runs a postinstall hook; dist/postinstall.js skips it unless the install is global and CI is unset.
The motion prewarm installs from a copied lockfile with dependency scripts disabled and checks fetched files against SHA-256 pins.
Agent and plugin setup runs through the explicit litcodex install command and targets the Codex home.
The cited Codex probe only lists plugins to check whether LitCodex is installed.
The flagged test, document, and page-probe files show ordinary test, DOCX, and DOM-inspection behavior.
Evidence against
Behavioral surface
SourceChild process · Crypto · Dynamic require · Environment variables · Filesystem · Network · Shell · Web socket
Supply chainHigh entropy strings · URL strings
ManifestNo manifest risk signals triggered.
Scanned 322 files · 2.51 MB of source
11 external domains referenced
127.0.0.1
api.typesafe.ai
biomejs.dev
clangd.llvm.org
doi.org
github.com
lit.stage
raw.githubusercontent.com
registry.npmjs.org
www.apache.org
www.w3.org
Findings
2 critical · 4 high · 7 medium · 6 low
Automated scan signals; the review checked them and cleared this version. Where they are, most severe file first. Code opens a file's flagged lines, with their evidence, below.