The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from platforms like X (Twitter), Reddit, and YouTube to generate blog content. \n
Ingestion points: External URLs and summaries are stored in sources json files before being processed by LLMs.\n
Boundary markers: The writing instructions for Gemini 3.8 Flash include strict prohibitions against inventing facts and require every claim to be linked to a specific source entry.\n
Capability inventory: The skill can write to the local file system, execute local migration and review scripts, and trigger deployments via agent-fleet.\n
Sanitization: A mandatory review phase is performed by Claude Sonnet, which is tasked with cross-referencing all statements in the draft against the original source data to prevent hallucinations or the introduction of unverified information.\n- [COMMAND_EXECUTION]: The skill utilizes an orchestration tool (fleet-go) to manage tasks across different AI agents and runs local scripts for SEO analysis, image management, and repository migrations.\n- [EXTERNAL_DOWNLOADS]: The workflow involves fetching data from various external social media and documentation platforms for research and communicates with multiple external LLM providers.

blog