AI Security Review
scanned 3d ago · by lpm-firewall-aiNo confirmed malicious attack surface was established. The main risk is an install-time cosmetic mutation of a dependency OAuth callback page, but source inspection shows no token capture, exfiltration, or hidden execution path.
Decision evidence
public snapshot- package.json defines postinstall: node scripts/brand-oauth-page.mjs
- scripts/brand-oauth-page.mjs resolves @earendil-works/pi-ai/oauth and rewrites its utils/oauth/oauth-page.js with bundled logo/favicon/title
- templates/workspace/AGENTS.md seeds AI-agent workspace behavior on onboarding, including memory and proactive heartbeat guidance
- postinstall script is readable, cosmetic, idempotent, catches errors, and performs no network or credential reads
- brigade.mjs only validates Node version, filters warnings, enables compile cache, and imports dist/entry.js
- dist/security/injection-patterns.js contains invisible Unicode only inside a regex intended to detect prompt-injection payloads
- workspace persona files are created by user-invoked onboarding with wx/no-clobber behavior in dist/workspace/bootstrap.js
- network and token handling observed is package-aligned for OAuth, providers, chat channels, local Convex, and user-configured integrations
- no evidence of credential harvesting, exfiltration, persistence outside declared app state, destructive install-time behavior, or hidden payload execution
Source & flagged code
22 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgThis package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
dist/ui/onboarding.jsView on unpkgPackage source references child process execution.
dist/ui/onboarding.jsView on unpkg · L14Source writes installer persistence such as shell profile or service configuration.
dist/core/daemon/systemd.jsView on unpkg · L8Package source references a known benign dynamic code generation pattern.
dist/agents/org/pride-image.jsView on unpkg · L90Package source references dynamic require/import behavior.
brigade.mjsView on unpkg · L54Package source references weak cryptographic algorithms.
dist/agents/tool-loop-detector.jsView on unpkg · L1A single source file combines environment access, network access, and code or shell execution; review context before blocking.
scripts/convex-dev.mjsView on unpkg · L16Source reaches cloud instance metadata or link-local credential endpoints.
dist/infra/net/fetch-guard.jsView on unpkg · L4Source contains bidi control or invisible Unicode characters associated with Trojan Source attacks.
dist/security/injection-patterns.jsView on unpkg · L58Package source invokes a package manager install command at runtime.
scripts/convex-push.mjsView on unpkg · L75Package ships non-JavaScript build or shell helper files.
skills/video-frames/scripts/frame.shView on unpkgPackage ships high-entropy non-source blobs.
scripts/assets/brigade-favicon.icoView on unpkgPackage contains source files above the static scanner size ceiling.
dist/ui/brand-frames-cli.jsView on unpkgPackage contains an oversized executable-looking CLI entrypoint.
dist/ui/brand-frames-cli.jsView on unpkgHardcoded password in dist/agents/channels/bluebubbles/account-config.d.ts
dist/agents/channels/bluebubbles/account-config.d.tsView on unpkg · L28Hardcoded password in dist/agents/channels/bluebubbles/account-config.d.ts
dist/agents/channels/bluebubbles/account-config.d.tsView on unpkg · L34Hardcoded password in dist/agents/channels/bluebubbles/account-config.d.ts
dist/agents/channels/bluebubbles/account-config.d.tsView on unpkg · L35Hardcoded password in dist/agents/channels/bluebubbles/account-config.js
dist/agents/channels/bluebubbles/account-config.jsView on unpkg · L28Hardcoded password in dist/agents/channels/bluebubbles/account-config.js
dist/agents/channels/bluebubbles/account-config.jsView on unpkg · L34Hardcoded password in dist/agents/channels/bluebubbles/account-config.js
dist/agents/channels/bluebubbles/account-config.jsView on unpkg · L35