Writing Skills

This is the Onmax adaptation of Matt Pocock's writing-great-skills reference. For non-trivial edits, load GLOSSARY.md and use its vocabulary for predictability, invocation, information hierarchy, and failure modes.

Process

1. Define one job - identify the task, trigger, and boundary. If the request needs several jobs, split it into several composable skills.

2. Challenge the shape - before drafting, compare the request against the Quality Bar and glossary. If it would create a broad lifecycle skill, harness-specific wrapper, secret-handling shortcut, or multi-purpose catch-all, say so and propose the smaller composable shape.

3. Draft the skill - create:

   • SKILL.md with concise instructions
   • Additional reference files for rarely needed details
   • Utility scripts only for deterministic repeatable operations

4. Review the shape - check that the skill is concise, responsible for one thing, composable, progressively disclosed, harness-agnostic, documented, portable, and secure.

5. Validate direction - before finalizing a new or materially changed skill, run validate-direction on the emerging skill design when the skill changes agent behavior, routing, permissions, workflow sequencing, or cross-skill coordination. Carry the verdict into the final edit before declaring the skill done.