What problem does it solve? Coordinating large backlogs of pending development tasks manually leads to merge conflicts, lost progress on crashes, polluted git history, and inconsistent execution. This Skill orchestrates a disciplined multi-agent loop that processes task queues from .ai-memory/plans/pending/ with file-collision locking, crash recovery, and enforced commit hygiene. ## Core Features & Use Cases - Batched Multi-Agent Execution: Runs exactly 3 concurrent sub-agents, each assigned small disjoint micro-task chunks tracked in a locking matrix to prevent git merge conflicts. - Crash Recovery & State Tracking: Logs agent state to .ai-memory/temp/ and .ai-memory/temp-agents/, applies a 3-strike rollback policy with git checkout, and resumes from failure points. - Atomic Commits & CI/CD Discipline: Consolidates all changes into single atomic commits, bans local test/build execution (deferred to CI/CD), and monitors pipelines via GitMap Pipeline-AI with ETA-based waiting. - Use Case: A repository has 40 pending task files in .ai-memory/plans/pending/. Invoke this Skill to autonomously loop through all of them—spawning agents, committing chunks, consolidating completed plans—until the queue is empty. ## Quick Start Ask the agent to execute all pending tasks from the plans folder using the batched loop, and say "continue" to resume the loop until the queue is empty.
