Show HN: Sigil – compile a SKILL.md the model MUST run, not "might" (github.com)

Sigil has introduced a groundbreaking skill compiler that transforms skills written in markdown (SKILL.md) into structured agent harnesses capable of executing specific tasks with high fidelity. Unlike traditional models that interpret prompts, potentially skipping steps or misinterpreting instructions, Sigil ensures that every mandatory action is rigorously enforced. The compiler generates a typed program, making it impossible for weaker models to ignore critical steps. This shift fundamentally alters how tasks are executed, placing the power of reliability and control directly within the structure established by Sigil, regardless of the model's capabilities. The significance of Sigil lies in its ability to synthesize a clear pathway from skill specification to execution. It creates an analyzable intermediary, the Agent Graph IR (AG-IR), which serves as both a visualization tool and a means to verify the integrity of the task flow. Each compiled skill has gates for verification and rules clearly outlined, leading to improved performance even when using smaller or local models. This marks a substantial leap forward in application reliability and program control in AI/ML, as it allows developers to build and execute skills with guaranteed adherence to specified requirements, drastically reducing the potential for errors.