SKILL.make, a new specification that rewrites Agent Skills in Makefile style, replaces informal prose with a dependency‑driven execution graph. By structuring skills as targets, recipes, and variables, the format turns a SKILL.md file into a reproducible DAG that agents can resolve automatically. The approach keeps the logic close to familiar build tools for developers daily.

The token‑efficiency claim is noteworthy: the compact syntax shrinks a typical skill file by roughly 15%, and manual tuning can cut context size over 30%. Smaller files reduce prompt costs and fit tighter into model windows, while the automatic DAG solver eliminates the need for LLMs to guess the next step within production pipelines today.

A real‑world test on the “Skills for Real Engineers” set shows SKILL.make files outperform their markdown counterparts, cutting overall size from 66,394 to 56,387 tokens—a 15% drop. The format also supports modular composition, allowing skills to call targets across files, and enables audit trails via Git history and invocation stats for continuous integration and deployment efforts.

The specification ships under an MIT license and targets compatibility with most Agent Harness implementations. By aligning skill definitions with a proven build‑system model, developers gain clearer execution logic, easier debugging, and a foundation for evolution engineering. SKILL.make offers a practical, lightweight alternative to prose‑based skill authoring for robust automation workflows in AI projects today.