EvanFlow – A TDD driven feedback loop for Claude Code

EvanFlow introduces a meticulously crafted, TDD-driven feedback loop designed to bring discipline and guardrails to Claude Code agentic development. It addresses common LLM coding agent failure modes through explicit checkpoints, vertical-slice TDD, and human oversight. Hacker News found it intriguing, sparking discussions on its methodology, the practicalities of AI-assisted development, and a dash of playful critique regarding its nomenclature.

Score

Comments

Highest Rank

on Front Page

First Seen

Apr 27, 3:00 AM

Last Seen

Apr 27, 11:00 AM

Rank Over Time

The Lowdown

EvanFlow presents an opinionated, TDD-driven iterative feedback loop tailored for software development using Claude Code. Developed by evanklem2004, this system comprises 16 cohesive skills and two custom subagents, guiding an idea from brainstorm through execution and iteration with strict checkpoints to maintain human control.

Core Workflow: The primary loop follows a brainstorm -> plan -> execute -> tdd -> iterate -> STOP sequence, ensuring developers approve each major stage.
Conductor, Not Autopilot: A central tenet is human agency, with explicit guardrails preventing auto-commits, auto-staging, or automated pull requests; every critical Git operation requires explicit user direction.
Key Disciplines: The system integrates vertical-slice TDD (one failing test -> minimal implementation -> repeat), embedded "grilling" at decision points to stress-test designs, and an "iterate-until-clean" phase (capped at five rounds) that includes checks against common failure modes and visual UI verification.
Advanced Orchestration: For plans with multiple independent units, EvanFlow forks into a parallel coder/overseer orchestration, complete with integration tests to enforce cohesion.
Hard Rules: The system bakes in rules derived from industry research on agentic coding failures, such as never inventing values, explicit assertion-correctness warnings for tests, context drift monitoring, and a "Five Failure Modes" checklist.
Comprehensive Skill Set: Beyond the core loop, it offers special-purpose skills for tasks like debugging, architecture improvement, interface design, glossary extraction, and code review.
Installation & Customization: EvanFlow supports multiple installation paths (Claude Code plugin marketplace, npx skills, or manual setup) and is designed to be easily customized to project-specific conventions.

Ultimately, EvanFlow aims to instill structured rigor and mitigate the inherent risks of autonomous LLM coding, making AI-assisted development more reliable and controllable.

The Gossip

Nomenclature Notoriety

The naming of the tool, "EvanFlow," sparked a notable side discussion. Some commenters questioned the perceived self-absorbed nature of naming software after oneself, while others made humorous connections to the "Evenflo" brand of baby products. The author, evanklem2004, clarified in their initial comment that the name reflects an "opinionated Claude Code development flow based on evidence based practices and what has been working for me while developing professional code," suggesting a personal yet proven methodology.

TDD Technique Tussle

Commenters engaged with EvanFlow's Test-Driven Development (TDD) approach. There was discussion comparing EvanFlow's vertical-slice TDD to other agentic development tools, with some noting that TDD is already a well-integrated feature in alternatives like `superpowers`. A specific point of critique concerned the apparent absence of a dedicated "refactor" step in EvanFlow's TDD loop. One commenter also posited a concept of "horizontal" TDD for LLMs, aiming to establish invariants before coding, as an alternative to vertical-slice.

Agentic Artistry Appraisals

The broader implications and perceived maturity of AI coding agents were discussed. Questions arose regarding EvanFlow's ability to navigate "dumb zones" during iterative loops, a common challenge in agentic development. A more direct criticism, stating a reluctance to use a product from someone perceived as an "undergrad," was quickly met with pushback, highlighting the debate around the validity of dismissing tools based on the author's perceived experience level. The author, evanklem2004, proactively addressed many of these concerns by detailing the evidence-based practices and guardrails integrated into EvanFlow's design to mitigate common LLM failure modes.