deep-think

SkillProductivity

Breaks a complicated problem into smaller reasoning steps and evaluates multiple possible approaches before recommending one. Use for genuinely hard, multi-faceted problems, not routine tasks.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the deep-think skill

What this skill tells your AI

The instructions your AI receives, as published by codebygarv/ai-skills in skills/reasoning/deep-think/SKILL.md and read by ahel’s review.

Purpose

Slow down on a hard problem: decompose it into its component sub-problems, generate more than one candidate approach, reason through each honestly, and only then converge on a recommendation. This exists to counteract jumping to the first plausible-sounding solution on problems that deserve better.

When to Use

  • The problem has multiple interacting constraints (technical, product, resourcing) and the right approach isn't obvious.
  • A first-instinct answer feels too easy for how hard the problem sounded.
  • The user explicitly asks to "think deeply," "reason through," or "really think this through" on something.

Do not use for simple, well-understood tasks — decomposing a one-step problem into stages wastes the user's time.

What to Analyze / Do

  1. Decompose the problem into its real sub-problems — what has to be true, in what order, for any solution to work.
  2. Identify constraints that any valid approach must satisfy (explicit ones stated by the user, and implicit ones like existing architecture, team size, timeline).
  3. Generate at least two genuinely different candidate approaches — not one approach and a token strawman.
  4. Reason through each candidate against the constraints and sub-problems, noting where each one is strong and where it breaks down.
  5. Converge: pick a recommendation, and state explicitly why the alternatives were not chosen.

Output Format

  • Short problem restatement, including the sub-problems identified.
  • Constraints list.
  • Each candidate approach as its own section: description → strengths → weaknesses.
  • Final recommendation with explicit reasoning for why it beats the alternatives.

Avoid

  • Presenting only one approach dressed up as if several were considered.
  • Decomposing trivial problems just to look thorough.
  • A recommendation that ignores a constraint identified earlier in the same analysis.

Signals

GitHub stars
25
Forks
1
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
deep-think
Source
github.com/codebygarv/ai-skills