The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external URLs or provided HTML, creating a surface for indirect prompt injection. Ingestion points: URL and page content (HTML) provided by the user or fetched from the web in SKILL.md. Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions found within the analyzed page's content. Capability inventory: The skill utilizes a local execution tool (claude-seo) to run metadata analysis scripts as seen in SKILL.md. Sanitization: No sanitization or escaping of the ingested page content is specified before it is processed.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a specific command-line tool to perform heuristic analysis on metadata. Evidence: The instruction to run '${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo' run metadata_template.py in SKILL.md.
[DYNAMIC_EXECUTION]: The skill constructs shell commands dynamically by interpolating variable content (page title and description) into the command arguments. Evidence: SKILL.md instructs the agent to populate --title '' and --description '' with content retrieved from the target website, which could lead to command injection if the strings contain shell metacharacters.

seo-page