Writing for agents

On this page 6

stacks-writing-for-agents · Engineering craft · model-invoked

The skill for writing skills, and for every other document an agent reads. It covers context pointers (why a weakly worded description is a variance bug), the information hierarchy, completion criteria, leading words, and the pruning discipline that keeps a document from silting up. Read it before you add anything to app/Skills/.

When to reach for it

  • A SKILL.md under app/Skills
  • storage/framework/defaults/ai/skills
  • The project AGENTS.md
  • A reference file a skill points at

Covers

context pointers, information hierarchy, completion criteria, leading words, pruning, the skill mechanics behind app/Skills, buddy setup:ai.

Inside the skill

The sections an agent reads once the skill loads.

  • Context pointers
  • The two loads
  • Information hierarchy
  • Steps and completion criteria
  • When to split
  • Leading words
  • Pruning
  • Before you finish

Supporting files

Reference and scripts the skill reaches for on demand, rather than loading up front.

Using it

This one is model-invoked. Your agent reaches for it on its own when the task matches, and you can also call it by name:

/stacks-writing-for-agents

Source: stacks-writing-for-agents/SKILL.md. Shadow it for one project with app/Skills/stacks-writing-for-agents/SKILL.md, then re-run buddy setup:ai. See Writing your own.