The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends and implements the installation of the 'ax' CLI tool via a shell script downloaded from 'https://ax.yusuke.run/install' and piped directly into the system shell.
[PRIVILEGE_ESCALATION]: The setup script ('scripts/setup.sh') contains logic to move the downloaded binary into system directories like '/usr/local/bin' using 'sudo', which is a high-privilege operation standard for global software installation.
[COMMAND_EXECUTION]: The skill includes Python, Node.js, and Bash wrappers that execute the 'ax' binary through shell commands to perform web discovery and data extraction tasks.
[EXTERNAL_DOWNLOADS]: The skill downloads binaries and installation scripts from 'ax.yusuke.run' and 'github.com/yusukebe', which are the official project sites for the 'ax' tool.
[INDIRECT_PROMPT_INJECTION]: The skill enables agents to fetch and process web content from arbitrary URLs, creating an attack surface where malicious data from external websites could attempt to influence the agent's behavior.
Ingestion points: Web content is ingested whenever the 'ax' command is used to fetch a URL (e.g., in 'SKILL.md' and 'tools/ax_tool.py').
Boundary markers: The tool supports token budgeting via the '--budget' flag and structural discovery using '--outline' to limit the amount of raw, potentially unsafe HTML returned to the agent.
Capability inventory: The skill provides capability for project scaffolding (filesystem writes) and CLI execution (subprocess calls in wrappers).
Sanitization: Employs a specific, restricted expression language for the '--where' filtering flag to prevent code injection within the tool's filtering logic.

ax