agentflywheel

Iteratively refine beads for ralph-tui execution through 4-6 polishing rounds until convergence, ensuring each bead fits one agent context window, has quality gates appended, uses correct bd CLI syntax, and follows schema-backend-UI dependency ordering. Use when user wants to polish beads for ralph, refine ralph tasks, iterate on beads before running ralph-tui, review bead quality for ralph execution, or check if beads are ralph-ready. Triggers on: polish beads ralph, refine beads for ralph, ralph bead review, check beads, are beads ready, polish for ralph-tui, iterate ralph beads.

Iteratively refine task breakdowns (beads, issues, stories, tickets) through 4-6 polishing rounds until convergence. Based on the 'Check N Times, Implement Once' principle. Use when user wants to polish beads, refine tasks, review task breakdown quality, improve issue definitions, iterate on a plan's task list, check task coverage, or deduplicate tasks. Triggers on: polish beads, refine tasks, polish tasks, review breakdown, check coverage, deduplicate issues, iterate tasks, polish issues, converge tasks.

Generate multiple independent implementation plans in parallel, then synthesize a best-of-all-worlds plan combining the strongest ideas from each. Use when planning a feature, designing architecture, creating an implementation strategy, or when the user says competing plans, parallel plans, plan shootout, multi-plan, best-of-all-worlds, or plan synthesis. This skill produces dramatically better plans than single-shot planning because different agent perspectives emphasize different aspects — one catches edge cases another misses, one proposes cleaner architecture while another has better error handling.

Enforce 85% planning before any implementation. Produce a comprehensive markdown plan covering architecture, data flow, error handling, edge cases, and security before writing code. Use when user wants to build a feature, implement something, start a new project, design architecture, scaffold an app, create a system, or begin any substantial coding work.

Structured divergent-convergent ideation pipeline for generating feature ideas for existing projects. Follows a specific 30 → 5 → 15 expansion/contraction pattern grounded in the current codebase to avoid both tunnel vision and unfocused brainstorming. Use this skill whenever the user wants to brainstorm features, generate ideas for a project, come up with new functionality, explore what to build next, think about improvements, wants a feature ideation session, mentions "idea wizard", asks "what should I build", says "feature ideas", or wants to figure out what to add to an existing codebase. Also use when user mentions wanting to adapt ideas from an external project or system into their own.

Three-mode code review system that catches different bug categories by alternating review perspectives. Use when user says "review", "code review", "check my code", "find bugs", "look for issues", "review my changes", "what did I miss", "anything wrong with this", "sanity check", "spot check", "audit this code", "review PR", "check for bugs", "is this correct", or any request to evaluate code quality. Also use after completing a feature implementation, before merging, or when the user seems unsure about correctness. Trigger aggressively - if there is any hint the user wants code evaluated, use this skill.