The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/install.sh script downloads an application installer (DMG) from https://cdn.ego.app/setup/macos/arm64/egolite.dmg (or x64 version).
[PRIVILEGE_ESCALATION]: The scripts/install.sh script uses sudo to perform administrative tasks, including removing Gatekeeper quarantine attributes via xattr, deleting existing application directories, moving files to the /Applications folder, and executing the macOS installer utility.
[COMMAND_EXECUTION]: The skill instructions guide the agent to execute a shell setup script and to run arbitrary Node.js code through the ego-browser CLI using a heredoc pattern.
[DYNAMIC_EXECUTION]: The skill uses the js() helper to evaluate code strings within the browser context and executes Node.js code strings via the ego-browser CLI.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted web data and returns it to the agent, creating a surface for indirect prompt injection. * Ingestion points: Web content is ingested via snapshotText() and various site-specific extraction scripts in the learnings/ directory (e.g., search-extract.js, timeline.js). * Boundary markers: Scraped web content is not delimited or marked as untrusted. * Capability inventory: The agent has access to the Bash tool for shell commands and can perform network requests via serverFetch and browserFetch. * Sanitization: No evidence of sanitization or validation of the scraped web content before it is passed to the agent.
[DATA_EXFILTRATION]: The skill is designed to inherit the user's active browser login state, which allows the agent to access authenticated websites and potentially sensitive user data.

ego-browser