What problem does it solve? Running Claude Code autonomously over long horizons is hard: context resets between invocations, parallel agents conflict, and naive loops waste tokens or stall on CI failures. This Skill provides proven patterns and architectures for building self-running development loops without human intervention. ## Core Features & Use Cases - Loop Pattern Spectrum: Six patterns from simple claude -p sequential pipelines and the NanoClaw REPL to infinite agent loops, Continuous Claude PR loops, and RFC-driven DAG orchestration (Ralphinho). - De-sloppify Pattern: A dedicated cleanup step after implementation that removes test slop and over-defensive code without constraining the implementer. - Merge Queue & Recovery: Dependency DAG execution with tiered quality pipelines, worktree isolation, and eviction-context recovery for merge conflicts. - Use Case: You need to add tests across a large codebase over several days. Use the Continuous Claude loop with --max-runs, a SHARED_TASK_NOTES.md context bridge, and automatic CI-failure repair to iterate and merge PRs unattended. ## Quick Start Ask the agent to set up a sequential claude -p pipeline that implements a feature with TDD, runs a de-sloppify cleanup pass, verifies the build, and commits.
