AgentZero

AdvisoryAudited by Static analysis on Apr 30, 2026.

Overview

No suspicious patterns detected.

Findings (0)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

If enabled, the skill can use the configured Gmail account to read Redfin alert metadata and drive listing discovery.

Why it was flagged

This is high-impact account access: the workflow depends on Gmail credentials or tokens stored locally. It is purpose-related, but sensitive enough that the user should explicitly review and constrain it.

Skill content
himalaya must be configured with an IMAP/SMTP account pointing to the user's Gmail inbox. The app password or OAuth token must be stored in the system keychain or himalaya config file.
Recommendation

Use the narrowest Gmail credential possible, confirm himalaya is configured only as intended, and enable the email scan only via a schedule or manual command you control.

What this means

A user may believe the browser automation never touches Gmail when the workflow actually describes opening Gmail briefly.

Why it was flagged

The same privacy section both says Gmail is not browsed and says Gmail is opened to search/click links, which could mislead users about browser session exposure.

Skill content
Browser automation: the openclaw browser is used exclusively to open Redfin listing URLs ... it does NOT browse Gmail or any other site. Gmail is only opened briefly to search for and click through to the Redfin listing link
Recommendation

Clarify exactly when Gmail is opened, what page content is inspected, whether cookies/session data are used, and what is logged or stored.

What this means

Listing data and search criteria may be processed by Anthropic during automatic triage.

Why it was flagged

The backend sends property details and user search profile descriptions to Anthropic Claude for triage. This is disclosed and purpose-aligned, but it means personal housing preferences leave the local app.

Skill content
const CLAUDE_API_URL: &str = "https://api.anthropic.com/v1/messages"; ... Fetch all search profiles ... Build a compact prompt from the parsed property fields.
Recommendation

Install only if you are comfortable sending those listing/search-profile details to Anthropic, and keep the API key scoped and revocable.

NoteMedium Confidence
ASI10: Rogue Agents
What this means

A cron job could keep scanning email/listings on a schedule after initial setup.

Why it was flagged

The skill is intended to run recurring automation. That fits a listing watcher, but persistent scheduled tasks should be explicitly approved and easy to disable.

Skill content
Ask your AI to install AgentZero and set up a daily cron job
Recommendation

Review the cron entry, choose the schedule yourself, and remove or disable it when you no longer want automated scans.

What this means

Running setup may download and execute dependency code from package ecosystems.

Why it was flagged

The setup relies on shell scripts and third-party Rust/npm dependencies. This is expected for the local backend/frontend, but it introduces normal supply-chain review requirements.

Skill content
scripts/run_backend.sh builds and starts the Rust/Axum backend ... scripts/run_frontend.sh runs `npm install && npm run dev` ... fetching Rust crates (cargo) and npm packages on first build.
Recommendation

Review the scripts and dependency files before running them, and install from a trusted checkout.