Skillification

Skillification is Garry Tan's term for converting a successful or repeated agent workflow into a durable skill file with triggers, instructions, edge cases, tests, and resolver wiring. The pattern turns one-off prompting into reusable infrastructure: when a workflow improves, every future run benefits from the fix.^{source: garry-tan-meta-meta-prompting-ai-agents-2026.md}

The recursive part is the meta-skill: Tan describes skillify as a skill that creates new skills. A manual book-mirror workflow became a reusable book-mirror skill; meeting prep became a meeting-prep skill; quality fixes such as fact-checking and cross-modal evaluation were baked into the skill so later runs inherited them.^{source: garry-tan-meta-meta-prompting-ai-agents-2026.md}

This is a practical mechanism for harness-engineering. Instead of asking which model is best, skillification asks which surrounding process should be captured so the model has less to rediscover. It also gives a concrete workflow for building a self-improving-knowledge-base: do real work, inspect output, extract the repeatable pattern, register it, and keep iterating.^{source: garry-tan-meta-meta-prompting-ai-agents-2026.md}

Shopify's shopify-river example shows a team-scale version of skillification: channels can pre-load the zones, skills, and instructions their teams need, written by the people closest to the work. Patterns discovered in public agent conversations can then spread to other teams instead of remaining private prompt craft.^{source: tobi-lutke-learning-shop-floor-river-2026.md}

Skillification

Resources