AI Security Review
scanned 8d ago · by lpm-firewall-aiNo confirmed malicious attack surface was established. The package is an agent/sandbox authoring facade with dangerous capabilities exposed as explicit user-invoked runtime tools.
Decision evidence
public snapshot- dist/agent.mjs can create persistent agent memory files when agent runtime is invoked
- dist/dist-Bmjy7TiE.mjs exposes MCP stdio spawning from user-supplied definitions
- dist/js-exec-N5KEZBH7-BYZdHawz.mjs contains a sandboxed JS execution tool with fetch/fs/child_process-like APIs
- package.json has no install/preinstall/postinstall lifecycle hooks and exports only dist entrypoints
- Network use is aligned to Keystroke/AI gateway clients, model catalogs, MCP execution, and OAuth flows
- Shell/process execution is exposed as explicit sandbox/MCP tooling, not import-time behavior
- File writes are scoped to agent memory, sandbox/workspace materialization, credentials, or user-invoked tools
- No credential harvesting or exfiltration path found beyond configured API tokens sent to configured platform endpoints
- Dynamic import in dist/config.mjs loads user project keystroke config by explicit loadConfig(cwd) call
Source & flagged code
6 flagged · loading sourcePackage source references child process execution.
dist/dist-jRbaylLr.mjsView on unpkg · L17378Package source references a known benign dynamic code generation pattern.
dist/chunk-TN7HHBQW-D_SK6w5n.cjsView on unpkg · L17Package source references dynamic require/import behavior.
dist/diff-MWJFIG7X-42sr7rDc.cjsView on unpkg · L1Source spawns a local helper that also contains network and dynamic execution context; review data flow before blocking.
dist/js-exec-N5KEZBH7-BYZdHawz.mjsView on unpkg · L40This package version adds a dangerous source file absent from the previous stored version.
dist/dist-C7ptcoHx.cjsView on unpkg