repeatedly-apply-skill

# Repeatedly Apply Skill Orchestrate N passes of any skill against a target. Each pass is delegated to a fresh subagent with a unique mission derived from the skill itself. ## Architecture: Orchestrator + Subagents **You are the ORCHESTRATOR.** You do NOT apply the skill yourself. You delegate each pass to a subagent via the `Agent` tool. This solves: - **Laziness**: Each subagent is fresh — no accumulated fatigue - **Context**: Each subagent gets full context window for its one mission - **Verification**: YOU verify the subagent's work after each pass - **Compaction**: The progress file tells you where to resume ## Step 0: Plan the Missions **Before pass 1**, read the target skill's SKILL.md (or understand the slash command's purpose). Then generate N missions that are specific to THAT skill. Do not use generic missions. **Example for `ui-polish`:** 1. Layout hierarchy and spacing 2. Color palette and contrast 3. Typography scale and weight 4. Interactive states (hover, focus, active, disabled) 5. Micro-animations and transitions 6. Empty/loading/error states 7. Responsive breakpoints 8. Dark mode consistency 9. Visual rhythm and alignment 10. Final screenshot comparison **Example for `ubs` (bug scanner):** 1. Null/undefined paths 2. Error handling gaps 3. Race conditions and async 4. Input validation boundaries 5. Security (injection, auth) 6. Resource leaks (connections, listeners) 7. Type safety holes 8. Logic errors 9. Edge cases (empty, max, negative) ...