What problem does it solve? Documents written for AI agents often produce inconsistent behavior because their wording, structure, and length are not engineered for how agents consume context. This Skill provides a reference framework for writing any agent-facing document (skills, AGENTS.md, CLAUDE.md, or pointer-reached docs) so the agent follows the same process every run. ## Core Features & Use Cases - Context pointer design: Rules for writing descriptions and pointer lines that reliably trigger the right material, with front-loaded leading words and one trigger per branch. - Information hierarchy: A ladder model (in-file steps, in-file reference, disclosed reference) plus progressive disclosure and co-location guidance to keep documents legible and prevent sprawl. - Completion criteria and pruning: Techniques for writing checkable, demanding step completion criteria, and for removing duplication, stale caches, no-op sentences, and negation-based instructions. - Use Case: When creating a new skill or editing CLAUDE.md, consult this reference to decide what to inline, what to disclose behind a pointer, and how to word the description so the agent fires it reliably. ## Quick Start Ask the agent to review your new skill or CLAUDE.md draft using the writing-for-agents guidance and suggest what to inline, disclose, or prune.
