Evidence for AI clean decision
- `server/src/index.mjs` exposes an authenticated WebSocket PTY that starts a local shell.
- `server/src/agent.mjs` runs the user-installed `claude` CLI.
- `server/src/transcript-sync.mjs` can pair a narrow `~/.claude/projects/<slug>` transcript directory when sharing is enabled.
- `server/src/index.mjs` posts telemetry/transcript metadata to the configured sync service.
Evidence against
- `package.json` has no `preinstall`, `install`, `postinstall`, or `prepare` hook; `prepublishOnly` only builds for publishing.
- `server/bin/wild-workspace.mjs` exposes shell, login, service, and update behavior through explicit CLI commands.
- `server/src/skills.mjs` only reads Claude skill/command directories; the flagged character is a BOM-tolerant frontmatter regex, not bidi control flow.
- `server/src/transcript-sync.mjs` limits syncing to a project-specific Claude transcript directory and documents sharing as the activation switch.
- No source evidence of credential harvesting, hidden exfiltration, remote payload execution, or foreign AI-agent configuration mutation.
Behavioral surface
SourceChild process · Crypto · Dynamic require · Environment variables · Filesystem · Network · Shell · Web socket
Supply chainHigh entropy strings · Minified · Obfuscated · URL strings
ManifestNo manifest risk signals triggered.
Scanned 75 files · 1.89 MB of source
13 external domains referenced
- 127.0.0.1
- api.anthropic.com
- api.z.ai
- claude.ai
- fb.me
- github.com
- lea.verou.me
- opensource.org
- reactjs.org
- registry.npmjs.org
- sync.venturewild.llc
- workspace.venturewild.llc
- www.w3.org