The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides installation scripts (install.sh and install.ps1) and instructions in the README.md that involve downloading the skill's source code directly from its GitHub repository at install time.
[REMOTE_CODE_EXECUTION]: The recommended installation method utilizes shell pipe patterns (curl | sh and irm | iex) to execute the remote installation scripts. While these scripts originate from the vendor's own repository on a well-known service, this pattern bypasses local review of the executable content before it runs on the user's system.
[INDIRECT_PROMPT_INJECTION]: The auditing procedure documented in references/15-audit.md enables the agent to interact with untrusted external data, forming a potential attack surface.
Ingestion points: The agent is instructed to fetch and analyze content from live external URLs provided by the user for auditing purposes.
Boundary markers: The auditing instructions do not include clear delimiters or specific directions to ignore instructions that might be embedded in the content of the target websites.
Capability inventory: The audit procedure involves significant capabilities, including executing arbitrary JavaScript within a browser context (eval) and running shell commands through the chrome-devtools-axi utility.
Sanitization: There is no evidence of filtering or sanitizing the content of the audited websites before the agent performs its evaluations, which could allow a malicious website to attempt to manipulate the agent's logic or behavior during the audit.
[COMMAND_EXECUTION]: The skill's primary auditing logic relies on the agent executing shell commands via the chrome-devtools-axi tool to measure performance and accessibility metrics in a browser.

awwwards