The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, specifically raw HTML, rendered browser DOM, and existing JSON-LD/Microdata blocks from third-party websites. While this constitutes an attack surface for indirect prompt injection, the skill mitigates this risk through a highly structured workflow that emphasizes verification and human-in-the-loop oversight.
Ingestion points: Website metadata, rendered HTML elements, and existing structured data blocks extracted during the inventory phase (Workflow Step 1).
Boundary markers: While no specific text delimiters are mandated, the instructions require the agent to visibly separate evidence classes (Semrush MCP data, page evidence, validator output, and user facts) in its output.
Capability inventory: The skill is restricted to read-only MCP tool calls for site audits and the generation of text-based JSON-LD drafts; it possesses no capabilities for autonomous file system modification, network exfiltration, or command execution.
Sanitization: The instructions provide strict requirements to validate all drafted code against 'visible page content' and 'verified facts,' and explicitly prohibit the agent from deploying code or inventing entity data to satisfy validators.
[COMMAND_EXECUTION]: The skill describes a tool-based workflow using specific Semrush MCP functions (e.g., projects, site_audit, issue_details). These are used for data retrieval only and do not involve the execution of arbitrary shell commands or system-level operations.
[CREDENTIALS_SAFE]: No hardcoded credentials, API keys, or sensitive environment variable paths were detected. The skill instructions properly direct the agent to use project identifiers provided by the platform context.

schema-markup