reasoning-skills

Use when facing a complex, multi-part problem that would benefit from being broken into independent and dependent sub-questions before answering, rather than reasoned about as one tangled whole. Trigger on "atom of thought", "AoT", any request to "break down" or "break this down" (a problem, question, or decision), any request to reason or think "step by step", "decompose this problem", "atomic decomposition", or any complex analytical, planning, or reasoning task where a direct one-shot answer risks missing a sub-part or an implicit dependency between parts.

Run an engineering postmortem on a just-finished task or session — what was learned, where time was wasted, and the concrete fixes so it goes faster next time. Reconstructs the facts, hands them to ONE well-briefed reviewer subagent, and returns a root-cause analysis, a minimum-round-trip replay, and proposed edits to the skill/config/code that caused the friction. Use when the user says "postmortem", "retro", "what did we learn", "how could that have gone better", "how do we do this more efficiently next time", or asks to review how a piece of work went.

Use when you've just completed work and the user wants an honest audit of it — confidence levels, shortcuts taken, unverified assumptions, and known weaknesses. Triggers on: "confess", "be honest about this", "what are you unsure about", "what did you cut corners on", "what's the weakest part", "audit your own work", "what assumptions did you make", or after any significant output where the user wants unvarnished reflection rather than reassurance.