The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites (metadata, titles, descriptions) and the Lytics content table during the assessment phase. This data is processed to create taxonomies and execution plans. Malicious content embedded in website metadata could potentially influence the agent's logic during the advisement phase. \n
Ingestion points: site-assessment.md (live website fetch), assessment.md (content scan via API). \n
Boundary markers: The instructions lack explicit boundary markers or delimiters for interpolating external content into prompts. \n
Capability inventory: The skill can perform API write operations and generate/execute local Python scripts. \n
Sanitization: Relies on JSON structure and URL encoding, with no specific sanitization for natural language instructions embedded in data. \n- [DYNAMIC_EXECUTION]: The execution.md file defines a process for generating a local Python 3 script (content_exec_*.py) to automate approved API actions. While the instructions specify strict guardrails—such as an HTTP method allowlist, a superset check for updates, and a prohibition on DELETE calls—the generation and execution of code at runtime is a dynamic execution pattern. \n- [COMMAND_EXECUTION]: The skill utilizes system commands like curl and python3 to interact with the Lytics API and perform site assessments. It provides specific shell snippets for authentication management and multi-account synchronization tasks. \n- [DATA_EXFILTRATION]: The skill communicates with api.lytics.io, the vendor's official API endpoint, and fetches content from customer-defined URLs for site assessment. These operations are essential for the skill's purpose and follow established protocols for authenticated API access to vendor infrastructure.

lytics-content