Deep expertise in agentic coding and the LLM ecosystem.

Building workflow tools and utilities through hands-on agentic development, fully delegating to coding agents, addressing real-world challenges as they arise, and transforming those insights into practical solutions. Also working on AI-native workflows beyond software development.

Through ongoing personal projects, delivering tools grounded in actual practice rather than speculation.

Sharing learnings through writing.

Latest: I Never Asked My AI Team to Build All of It

🛠 Agentic Development Frameworks

• claude-code-workflows - Keeps Claude Code focused on the result you agreed on, even when complex work uncovers tempting side issues
• codex-workflows - Keeps Codex from turning a focused request into extra features, settings, and complexity while still delivering what you asked for
• grok-workflows - Use claude-code-workflows on Grok without maintaining a separate copy of its design, implementation, and review workflows
• galley - Makes lower-cost or specialized coding models practical by pairing them with independent supervision and review-ready PRs
• ai-coding-project-boilerplate - Start a TypeScript project with a Claude Code setup your team can version, share, and use from design through verification
• claude-code-discover - Helps you decide what to build by testing the assumptions that matter, then carries the evidence into an implementation-ready PRD
• linear-prism - Turns requirements into a small set of Linear tasks that people or coding agents can pick up and verify independently
• nautilus - Keeps product discovery context in your repo as you move from rough questions to prototypes and evidence-backed PRDs in Cursor, Codex, or OpenCode

🔌 MCP Servers

• mcp-local-rag - Search private documents without sending them to an embedding service, while still finding related ideas and exact technical terms
• mcp-image - Turns rough image requests into stronger prompts without losing your intent, then generates or edits the result
• sub-agents-mcp - Run reusable task-specific agents from any MCP client, using the coding backend you choose

🧰 Utilities

• agent-clinic - Tells you why a Claude Code or Codex session went wrong, and whether a change your agent proposes is more or less than the outcome needs
• codex-subagent-playbook - Helps Codex choose the right model for each subagent, wait for it to finish, and check the result before moving on
• github-profile-stats - Publish GitHub profile cards from your own repository, with no hosted renderer to depend on
• jev-reranker - Decide which of a search command's results reach your LLM: reorder them, drop the ones with no usable evidence, or keep only the passages that answer the question
• metronome - Claude Code plugin that catches shortcut-taking behavior and keeps Claude working step by step
• pr-review-skill - Gives every reviewer the same complete PR snapshot, then lets you approve findings before they are posted
• rashomon - Tests whether a prompt or skill actually improves agent behavior before you ship it
• sub-agents-skills - Reuse the same agent definitions across coding tools, including Antigravity, Claude Code, Codex, OpenCode, and Pi.

📋 Awesome Lists

• awesome-codex-workflows - Find Codex workflow projects whose real design is visible in code, configuration, or runtime behavior