decompose-skill
Decompose Skill
Decompose reusable responsibilities from one skill into independently invocable workflow skills and deliberate fresh-context handoffs; do not extract source-private steps, incidental helpers, or orchestration that has no credible use on its own.
Inputs
A. Target skill
A required repository-relative skill directory or SKILL.md path. Read the complete skill directory and identify the behavioral entry point to analyze. The target remains the public orchestrator unless the approved plan explicitly replaces it.
B. Decomposition controls
Optional user-supplied focus areas, proposed child names, naming constraints, configuration preferences, agent or subagent boundary preferences, or exclusions. In their absence, analyze the whole target, derive Agent Skills-compliant kebab-case names, and place child skills under the active repository's .agents/skills/ root through workflow-to-skill.
C. Repository context
Required repository instructions, skill-root conventions, relevant sibling skills, references, scripts, consumers, tests, validation commands, and available agent or subagent delegation mechanisms. This context is authoritative for local paths and behavior and supplies evidence for or against independent use.