What problem does it solve? Large development tasks often fail when handled in a single pass: specs get skipped, sub-agents hallucinate, coding guidelines are ignored, and work is left uncommitted. This Skill enforces a disciplined, multi-phase orchestration loop that plans, delegates, verifies, and ships complex tasks autonomously. ## Core Features & Use Cases - Spec-First Decomposition: Writes a detailed implementation spec into .ai-memory/plans/pending/ and breaks it into focused subtask files before any code is written. - N-Step Continuous Loop: Runs a configurable number of steps (default N=130), splitting the run between spec writing and execution with strict coding guideline enforcement. - Sub-Agent Management: Spawns lean, task-focused sub-agents, kills and rolls back failing ones, and enforces a 3-strike failure rule with logging. - Guardrails & Bans: Prohibits test running, build checking, per-file commits, and rapid CI/CD polling, deferring verification to CI/CD pipelines. - Use Case: Hand a large feature request (e.g., implementing a new standard UI alongside an existing modern UI) to the agent, which writes the spec, decomposes it into subtasks, executes them under strict naming/error-handling rules, and finishes with a single atomic commit and push. ## Quick Start Ask the agent to execute the parent task for your feature request using the execute-parent-task workflow and let it plan, decompose, and implement it end to end.
