What problem does it solve? Writing agent definitions that delegate correctly is hard: facts get duplicated between skills and agents, discovery steps repeat, gates lack proof, and descriptions fail to route. This Skill provides the rules for dividing knowledge between skills and agents, structuring agent files, and validating the finished result. ## Core Features & Use Cases - Skill/agent divide rules: Decides where each fact lives so knowledge is stated once, with criteria for when repeated discovery steps and gates justify a new agent. - Section-by-section authoring standards: Defines the twelve sections of an agent file (role, context_gathering, sources, decision, procedure, gate, done_when, and more) with concrete writing rules for each. - Proven tool forms reference: Ships a references file of verified command forms and behaviors for git, fd, rg, sd, mise, Nx, dotnet, ast-grep, pytest, ruff, biome, and MCP tools. - Use Case: When rebuilding an existing agent definition, apply the checks section to run ast-grep scans, claude plugin validate, and git diff against the earlier revision to confirm no commands or rows were dropped. ## Quick Start Use the agent-builder skill to review my agent definition file and check its sections, gates, and description against the standards.
