writing-for-agents
Organizes instructions an agent can apply predictably.
- Status
- Observed
- Last reviewed
- 2026-10-10
ID: writing-for-agents.
What it does
Distinguishes actions from references and decides what must be read immediately and what only when a condition requires it. Makes explicit the cases that trigger opening a skill or document and the completion criteria for each step. Removes duplication, ineffective instructions and conventions easily inferred from the environment. For skills, also considers descriptions, invocation modes and router boundaries.
When to use it
Use it when writing or editing SKILL.md, AGENTS.md, CLAUDE.md or operational documents agents must consult. It is particularly useful when an agent skips steps, opens too many references or cannot tell when a procedure is needed.
How to use it
Ask for it by name, or invoke it with /writing-for-agents in Claude Code or $writing-for-agents in Codex. Provide the document, audience, usage conditions and observed execution problems. Request verifiable criteria and one source of truth per rule. Keep common steps readable and move specific details behind references with clear conditions.
What it delivers
- Instructions with steps and completion criteria
- Conditional links to references
- Consistent description and invocation boundaries
Example
Review this routing skill: make explicit when it applies, which decisions it preserves and when each step is complete, without duplicating the method graph.
Executable instructions
The skill is installed from the source maintained by mattpocock; its license applies.
In the workflow
Use the router to choose this capability based on the required outcome.