AI Security Review
scanned 2h ago · by lpm-firewall-aiLPM treats this as warn-only first-party agent extension lifecycle risk. The package is a Twinkle/Lumine CLI with explicit commands for login, pulling/saving projects, assets, thumbnails, and SDK calls. The main unresolved firewall concern is user-invoked creation of first-party AGENTS.md/CLAUDE.md guidance in pulled workspaces, not install-time hijacking.
Decision evidence
public snapshot- lib/workspace.js writes AGENTS.md and CLAUDE.md into pulled workspaces when absent or marker-owned.
- lib/constants.js embeds Lumine agent instructions that steer Codex/Claude workflow for Twinkle projects.
- lib/auth.js stores scoped login tokens in ~/.twinkle/lumine-cli-auth.json after explicit login.
- lib/api.js and lib/http.js send authenticated requests to Twinkle API endpoints.
- package.json has no preinstall/install/postinstall lifecycle scripts.
- bin/lumine.js only invokes CLI main; behavior is user-command driven, not import-time.
- Network hosts are package-aligned defaults: api.twinkle.network, www.twin-kle.com, registry.npmjs.org.
- No eval/vm/Function, obfuscated payload, native binary loading, or remote code execution found.
- Workspace file writes validate paths stay inside the selected project directory.
- AGENTS.md/CLAUDE.md writes avoid clobbering customized files by requiring the Lumine marker.
Source & flagged code
4 flagged · loading sourcelib/workspace.js writes AGENTS.md and CLAUDE.md into pulled workspaces when absent or marker-owned.
lib/workspace.jsView on unpkglib/constants.js embeds Lumine agent instructions that steer Codex/Claude workflow for Twinkle projects.
lib/constants.jsView on unpkglib/auth.js stores scoped login tokens in ~/.twinkle/lumine-cli-auth.json after explicit login.
lib/auth.jsView on unpkglib/api.js and lib/http.js send authenticated requests to Twinkle API endpoints.
lib/api.jsView on unpkg