think_skill

Overview

This skill applies Reflective Thought Composition (RTC) to break down and analyze complex problems with a repeatable, structured reasoning pipeline. It guides the agent through restating the problem, generating options, critiquing assumptions, exploring alternatives, scoring choices, and delivering a clear recommendation. Use it to produce evidence-backed conclusions and transparent chains of thought.

How this skill works

The process enforces a six-step inspection for every problem: restate the prompt, ideate multiple approaches, reflect critically on weaknesses, expand orthogonally to find overlooked angles, score and rank options, then respond with a recommended path. Depth can be adjusted with a depth flag (d=1 to d=10) to vary brevity versus thoroughness. Each step shows concise, information-dense work so reasoning is auditable and actionable.

When to use it

• Solving multi-step engineering or design problems
• Comparing trade-offs between competing technical approaches
• Planning complex projects with interdependent risks
• Preparing for high-stakes decisions that need documented reasoning
• When asked to "think deeply" or show a chain of thought

Best practices

• Always start by restating the problem in your own words to confirm scope and constraints
• Generate at least three distinct approaches during ideation to avoid tunnel vision
• Explicitly list assumptions and potential failure modes in the reflection step
• Use orthogonal expansion to surface unconventional or cross-domain options
• Apply a simple, consistent scoring rubric during evaluation so comparisons are reproducible

Example use cases

• Designing a scalable architecture and comparing cost, latency, and operational complexity
• Choosing between algorithms by weighing accuracy, runtime, and dataset constraints
• Preparing a research plan or PhD-level literature review with prioritized experiments
• Planning product roadmaps that balance customer value, engineering effort, and risk
• Diagnosing ambiguous system failures by generating and testing multiple hypotheses