#ai#model-selection#software-development#cost-control#fleeting

2026 07 20

Model Selection for Matt Pocock Skills

Skill routine → model mapping with cost control strategy.

Skill Routine Active Model Rationale Cost Control Strategy
/wayfinder Kimi K3 Ingests the whole repository + logs; maps out foggy decisions cleanly. Lock OpenRouter session headers to keep the shared map warm in the cache.
/to-spec Kimi K3 Synthesizes complex context, determines testing seams, and handles vision inputs. Pure text-only features can be dropped down to K2.7 to save token caps.
/to-tickets / /tdd Kimi K2.7 Code Hyper-efficient, bounded implementation of granular feature blocks. Use local terminal caching plugins to retain context between coding iterations.

Key insight: K3 for ingestion and synthesis (context-heavy), K2.7 for implementation (bounded, repetitive). Session/cache warming matters more than raw model power once the map is built.

Related:

Edit on GitHub