The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (rendered_dom, raw_html, and sitemap data) to perform its SEO audit. This creates a surface where malicious instructions embedded in a target website's HTML could attempt to influence the agent's behavior.
Ingestion points: The skill ingests rendered_dom or raw_html from URLs scanned via WebFetch.
Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent ignores or sanitizes instructions found within the processed HTML content.
Capability inventory: The skill has access to tools including Bash, WebFetch, Read, Grep, and Glob.
Sanitization: The instructions do not define any sanitization, filtering, or validation steps for the external content before processing.
[COMMAND_EXECUTION]: The verification process involves executing a shell command via Bash that incorporates a dynamic URL parameter.
Evidence: The verification step uses node ${CLAUDE_SKILL_DIR}/../../scripts/link-graph.mjs --url <u>.
Risk: If the URL placeholder <u> is replaced with input that is not properly escaped, it could lead to command injection.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to external websites to retrieve their content for auditing.
Evidence: Usage of the WebFetch tool to retrieve page snapshots for analysis.
[DYNAMIC_EXECUTION]: The skill relies on executing a local Node.js script located at a relative path to perform link graph verification.
Evidence: node ${CLAUDE_SKILL_DIR}/../../scripts/link-graph.mjs is executed during the verification phase.

seo-internal-linking