explore

§ 01 · purpose

Purpose

When several valid approaches exist, explore produces a side-by-side comparison instead of jumping to one. Surface trade-offs explicitly so the design pass starts with the architectural shape already chosen.

§ 02 · when to use

When to use it

• There are multiple valid implementations and you cannot pick on intuition alone.
• The space includes external libraries — you need an ecosystem scan with links.
• You're a valid pipeline entry point on its own (feeds design or blueprint directly).

§ 03 · inputs

Inputs

§ 04 · outputs

Outputs

§ 05 · key steps

Key steps

1. 01 Generate 2–4 named candidates from three sources Why: Ecosystem scan via web-search-researcher for external libraries · design-space enumeration for first-principles shapes · user shortlist for already-named picks. Merging the three avoids missing whole categories of solution.
2. 02 Confirm candidates × dimensions checkpoint Why: Dimensions (approach-shape · precedent-fit · integration-risk · migration-cost · verification-cost · novelty) are confirmed *before* per-candidate fit agents are dispatched — keeps the analysis budget proportional to the relevant axes.
3. 03 Per-candidate fit dispatch in parallel Why: One agent per candidate, scoring against the locked dimensions. Running concurrently keeps the comparison balanced — no candidate gets stale context from prior reasoning.
4. 04 Synthesize a comparison matrix + recommendation Why: A matrix is scannable; a recommendation forces a position. Both produce a clean handoff to design/blueprint, which expects a chosen shape rather than a menu.

§ 06 · related skills

Related skills