The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard curl and grep commands to inspect HTTP headers, status codes, and redirect chains for a user-specified domain. It also includes a loop to process local files like old-paths.txt for batch redirect verification. These are typical behaviors for website auditing tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by fetching content from the target website's homepage, robots.txt, and sitemap.xml. This creates a surface for indirect prompt injection if the target site contains malicious instructions hidden in metadata or headers.
Ingestion points: The curl commands in sections 2 and 4 fetch data from the target website (SKILL.md).
Boundary markers: The skill uses grep to filter for specific strings (e.g., meta tags, titles), which limits the data the agent processes, though it does not use explicit prompt delimiters.
Capability inventory: The skill performs network reads (via curl) and generates a summary report based on the findings.
Sanitization: No specific prompt-level sanitization is present, but the use of focused grep patterns serves as a functional filter for extracted content.

launch-site-checklist