The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax in SKILL.md to execute jq, ls, and lsof at load time. These commands are used to identify the project's framework, dev scripts, and occupied network ports to provide discovery context to the agent.
[COMMAND_EXECUTION]: The skill manages a development server by executing shell commands such as nohup for startup and kill for termination. It also performs network polling using curl. Some of these commands incorporate variables derived from user input or project files, which could lead to command injection if not properly handled by the agent.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from $ARGUMENTS (routes, ports) and conversation history, including pasted DOM fragments that specify file paths and line numbers.
Boundary markers: No specific boundary markers or safety instructions are defined to separate untrusted user data from agent instructions within the prompts.
Capability inventory: The skill possesses capabilities for shell command execution (nohup, curl, kill), browser-based script execution (evaluate_script), and file system modification (applying edits to source code).
Sanitization: The skill lacks explicit sanitization or validation logic for external data before it is used to target file edits or parameterize shell commands.

verify-ui