silvesterdivas

auto-pilot-handoff

This skill should be used when the context budget hits RED zone, when the context-budget hook fires, or when it is time to save progress and create handoff files before starting a fresh conversation.

This skill should be used when the context window is filling up, when deciding whether to read a file or use grep, when planning work scope, when the user asks about "budget zones", "context budget", or "token management", or when you notice the conversation is getting long.

code-intelligence

This skill should be used when navigating an unfamiliar codebase, searching for symbol definitions or usages, finding related files, or when asked how to search code without reading entire files.

degradation-signs

This skill should be used when forgetting earlier context, looping on errors, producing inconsistent code, or when the user says "you already did that", "that's wrong", "you're hallucinating", or "context is degrading".

model-switching

This skill should be used when selecting a model for a task, when asking "which model should I use", "is this a Haiku or Sonnet task", "use haiku", or when delegating work to subagents and choosing a model tier.

prompt-caching

This skill should be used when working to reduce token cost, when asked "how do I cut token costs", "why is my session expensive", "make this cache-friendly", or when ordering work within a long session so the prompt cache stays warm.

thinking-control

This skill should be used when calibrating reasoning depth for a task, when deciding whether to think deeply or act immediately, when asked about "thinking depth", "over-thinking", or "under-thinking".