skills/citrolabs/ego-lite/ego-browser/Gen Agent Trust Hub

ego-browser

Fail

Audited by Gen Agent Trust Hub on Aug 11, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The scripts/install.sh script downloads an executable DMG file from an external server at https://cdn.ego.app. This introduces unverified binary code into the local environment.- [REMOTE_CODE_EXECUTION]: The installation script executes remote code by mounting the downloaded DMG and running the installer command or using open to launch the application. The ego-browser nodejs tool further enables the execution of arbitrary Node.js scripts via heredocs provided by the agent.- [COMMAND_EXECUTION]: The skill executes high-risk commands within scripts/install.sh, including using sudo to escalate privileges for file replacement and system-level installation. It also uses xattr -dr com.apple.quarantine to explicitly bypass OS-level security warnings and Gatekeeper protections for the downloaded binary.- [DATA_EXFILTRATION]: The skill is specifically designed to inherit and reuse the user's browser login state, providing programmatic access to private, authenticated web sessions. When combined with the serverFetch and browserFetch helpers, this allows for the exfiltration of session tokens and personal data. The uploadFile helper also provides a mechanism to transmit local files to remote endpoints.- [PROMPT_INJECTION]: The skill processes untrusted content from external websites (e.g., Google, X) through snapshotText and screenshots. This exposes the agent to indirect prompt injection, where an attacker could embed malicious instructions in a web page to manipulate the agent's behavior during automated browsing tasks.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 11, 2026, 01:21 AM
Security Audit — agent-trust-hub — ego-browser