The Agent Skills Directory

[DYNAMIC_EXECUTION]: The provided Electron driver skeleton (examples/electron.md) implements a REPL driver that accepts an eval command, which uses page.evaluate() to execute arbitrary JavaScript strings provided via standard input within the browser context of the launched application.
[PRIVILEGE_ESCALATION]: The skill's setup instructions for Electron apps (examples/electron.md) include apt-get install -y commands for installing system dependencies, which necessitates root or administrative privileges on the host system.
[INDIRECT_PROMPT_INJECTION]: The driver scripts and interaction patterns established by the skill (such as REPLs and piped commands) create a vulnerability surface for indirect prompt injection. If an agent processes untrusted data that influences the commands sent to these drivers, it could lead to unauthorized actions within the application environment. Ingestion points: Standard input reading in driver.mjs and piped input to chromium-cli. Boundary markers: None identified. Capability inventory: Arbitrary JavaScript execution, UI interaction (click, type, press), and screenshot generation. Sanitization: None.
[COMMAND_EXECUTION]: The skill guides the agent to use extensive shell command patterns for application lifecycle management, including background execution (&), terminal multiplexing with tmux (sessions, send-keys, capture-pane), and process management using kill and lsof.

run