The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8). It instructs the agent to ingest and analyze untrusted data from external websites (raw HTML, rendered DOM, and text). It lacks boundary markers or 'ignore embedded instructions' warnings, creating a risk that a malicious website could contain instructions that the agent might inadvertently follow during the audit process.
Ingestion points: External URLs provided by the user for auditing (e.g., 'production URL, a preview deploy, a competitor's site').
Boundary markers: Absent. The instructions do not provide delimiters for the external content or tell the agent to treat it strictly as data.
Capability inventory: The skill utilizes browser control (viewport management, screenshots, console reading, network header inspection, User-Agent overriding) and shell command execution (curl).
Sanitization: Absent. There are no instructions to sanitize or validate the content fetched from the remote source before analysis.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell-level operations, specifically using curl to fetch raw HTML and driving browser tools (like headless Chrome) to perform technical audits. While these actions are aligned with the skill's stated purpose, they represent a high-privilege interaction with the host environment and network.
[EXTERNAL_DOWNLOADS]: The skill's core functionality involves fetching content from remote, untrusted URLs. It specifically recommends spoofing User-Agents (e.g., GPTBot, Googlebot) and fetching /llms.txt files, which involves multiple outbound network requests to user-specified targets.

tanner-website-review