---
name: thinking
description: "Decompose large, ambiguous, or blocked work into independent, testable units with explicit assumptions and ownership. Use before planning or dispatching work that cannot be solved safely as one step."
---

# thinking — decomposing a large or ambiguous problem into solvable units

Carried through with no stripping needed: the source contains zero references to any
concrete tool, and the discipline it states is already harness-neutral — decomposition,
stated assumptions, and delegation to subordinate execution units are concepts every
harness in `RECONCILIATION.md` shares, only the dispatch mechanism differs per harness.

## Problem solving

Every problem, no matter how large, is a unique combination of smaller ones — there is no
mountain to move, only many small pieces. If that premise doesn't hold, the work isn't
ready to start. To move all of them, move one first; if the smallest unit can't be moved,
the largest one can't either. At the working-agent scale: each unit is one concrete task
toward a specific outcome, and rather than moving the whole mountain alone, lean on
subordinate execution units and whatever fan-out the running harness provides. Even a
complete plan can't anticipate every possibility execution will surface — a redo is
inevitable sometimes; the actual skill is recognizing when one is needed and executing it
precisely, not avoiding the need for one.

## Breaking it down

1. **Name it before solving it.** A problem that can't be stated precisely can't be solved
   precisely either.
2. **State assumptions explicitly.** Every real problem carries something unknowable; name
   what's being assumed rather than letting it hide.
3. **Think in shapes.** Once the problem and its assumptions are stated, a structural
   sketch of the problem space is often the fastest way to confirm it's been captured
   fully, not just described.
4. **Solve it.** No fixed sequence covers every problem, but taking the moment to consider
   the actual environment first is usually the difference between a real fix and a
   plausible-looking one.