AI Security Review
scanned 3d ago · by lpm-firewall-aiNo confirmed malicious attack surface. The package is an AI-agent project scaffolding CLI with broad user-invoked capabilities: downloads updates, writes agent/MCP config, runs local project graph modules, and has optional remote collaboration/telemetry paths.
Decision evidence
public snapshot- Collab feature can upload local knowledge graph rows to a remote Turso DB via https://agentic-collab.adrianlpz-game.workers.dev when AKDD_COLLAB_ENABLED=1 and user runs collab commands.
- init/update fetch and execute/copy code from GitHub main rather than pinned npm contents, then write AI-agent control files (CLAUDE.md/.cursor).
- No install-time malicious action: package.json prepare only chmods bin/akdd.js; prepublishOnly is publisher-side.
- Main CLI import path only parses modules and dispatches commands; no import-time exfiltration seen.
- Network calls are visible, package-aligned features: GitHub update/install, optional collab sync, optional Claude/Voyage/telemetry integrations.
- No credential harvesting loop, broad env dump, hidden binary, obfuscation, or destructive persistence found.
- AI-agent instruction files are explicit product artifacts copied on akdd init/update, not unconsented npm lifecycle mutation.
Source & flagged code
5 flagged · loading sourceThis package version adds a dangerous source file absent from the previous stored version.
bin/akdd.jsView on unpkgPackage source references dynamic require/import behavior.
bin/akdd.jsView on unpkg · L3Package source invokes a package manager install command at runtime.
src/init.jsView on unpkg · L216