| name | learn-decompose |
| description | Step 2 of the Specificity Method. Break a specific "I don't know X" into its smallest moving parts, mark each piece known or unknown, and surface hidden unknowns. Use after the gap is articulated, when a topic is still a black box, or as step 2 of learn-anything. The unknown pieces become the real questions to research. |
Step 2 — Decompose into Pieces
The act of listing the work. You can't research a black box, but you can research a list.
"The thing" → decompose → piece 1 ✓ / piece 2 ✗ / piece 3 ✓ / piece 4 ✗ ...
How
- List the parts. Break X into the smallest fundamental components. For "Claude Code subagents": offer shaping, architecture, what a subagent is, forks, context-window savings, skill chaining, tool access, verification — each its own piece.
- Mark every piece known ✓ or unknown ✗ in
learn-state.md under ## Pieces.
- Surface the invisible. Decomposing makes unknowns you didn't even know existed visible. The first articulated gap is usually layered over several deeper questions — list those too.
- The ✗ pieces are your real questions. Research those, not the original vague topic.
Lists are tractable; fog is not. Every ✗ is a concrete question to hand to Step 3.
Next: hand off to learn-verify with the list of ✗ pieces.
