← ClaudeAtlas

agent-document-designlisted

Use when creating or editing Skills, agent instruction files, or referenced documents so triggers, steps, disclosure, completion criteria, and maintenance remain predictable.
sandbaseai/workbuddy-skill · ★ 2 · AI & Automation · score 81
Install: claude install-skill sandbaseai/workbuddy-skill
# Agent Document Design Use this Skill when an Agent-facing document is created, edited, split, linked, or retired. A good document causes the Agent to follow a predictable process on every relevant run; it does not merely produce the same output once. Design the trigger, information path, steps, references, and completion gate together. ## Design context pointers A context pointer is a reference that names material outside the current context and states when to load it. A Skill description and a line in an instruction file are both pointers. The wording controls invocation reliability, not just the target file. For each pointer: - put the leading trigger word first; - name the material's purpose and every genuinely distinct branch that should load it; - use one trigger per branch and remove synonym repetition; - avoid repeating identity already carried by the target; - ensure a required reference cannot be reached only through a vague pointer. Treat pointer text as an always-loaded context budget. Inline essential material when sharpening the pointer cannot make invocation reliable; disclose branch-specific detail when loading it on every run would distract or waste attention. ## Protect the information hierarchy Organize material by how immediately the Agent needs it: 1. **In-file steps**: ordered actions the Agent must perform now. 2. **In-file reference**: definitions or rules consulted while following those steps. 3. **Disclosed reference**: branch-specific mate