fpf-strategize - SKILL.md Agent Skill

name: fpf-strategize description: Bet on a method family — choose WHICH CLASS of approach to explore, not which specific solution. First-class creative act. MUST invoke for T4 architectural work before variant generation. argument-hint: "[problem-or-domain]"

What this skill IS

This is strategic betting, not variant selection. You are choosing which class of approach to bet on — scoping the search space for variant generation. "We'll explore variants within [family X] because [reason]."

Strategizing ≠ SoTA survey. SoTA maps the landscape; strategizing bets on a region. Strategizing ≠ variant selection. Strategy scopes the family; selection picks from generated variants.

Prerequisites

PROB-* or ANOM-* exists with acceptance spec
SOTA-* exists (if missing: invoke /fpf-sota first)

Output

.fpf/decisions/STRAT-${CLAUDE_SESSION_ID}--<slug>.md

Constraints (quality bar)

C1: ≥2 candidate method families evaluated
C2: Each family assessed for admissibility against PROB-* constraints
C3: Primary bet stated with rationale (why this family over others)
C4: Variant generation axes defined (what will differ between variants within this family)
C5: Invalidation conditions stated (what would cause a pivot to different family)
C6: Links to PROB-* and SOTA-*
C7: Scaling-Law Lens (SLL): for each family, state how it scales with more resources (compute/data/time). Minimum two measurement points (Scale-Probe) to claim a slope. Elasticity class: growing | plateau | declining. At comparable budgets, prefer the family with better scaling slopes (BLP — Bitter Lesson Preference).

Format

# Strategy Card
- **ID:** STRAT-...  **Problem:** PROB-...  **SoTA:** SOTA-...
- **valid_until:** YYYY-MM-DD

## Candidate families
| # | Family | Admissible? | Scaling (SLL) | Key risks |
| F1 | ... | Y/N | growing/plateau/declining | ... |
| F2 | ... | Y/N | growing/plateau/declining | ... |

## Bet
Primary: [family]. Rationale: ...

## Variant generation axes
(diversity dimensions within chosen family)

## Invalidation conditions
(what would cause a pivot)

## valid_until
(when does this strategy bet go stale? Tech landscape changes, new SoTA emerges)