EvanFlow is a TDD-driven iterative development workflow built on top of Claude Code, consisting of 16 skills and 2 custom subagents. It guides a software idea through brainstorm → plan → execute → TDD → iterate with explicit human checkpoints at each stage. Key design principles include never auto-committing, never inventing values, and enforcing a Five Failure Modes checklist (hallucinated actions, scope creep, cascading errors, context drift, tool misuse). For parallelizable work, it forks into a coder/overseer orchestration pattern where read-only overseer subagents review but cannot modify code. A bundled git guardrail hook blocks destructive git operations. The project synthesizes research on agentic coding failure modes, including findings that 62% of LLM-generated test assertions are incorrect and that ~65% of enterprise AI coding failures trace to context drift.
Table of contents
Quick InstallWhat Makes It a Feedback LoopHard Rules Baked Into the LoopThe Skill SetHard Rules (apply to every skill)RequirementsInstallationCustomizationHow EvanFlow Works End-to-EndRepository StructureCreditsLicenseContributingSort: