Agentic Workflows

A collection of all my agentic workflows, skills, prompts, etc.

🛠️ Skills Included

• autodev: Universal Autonomous Feature Development Protocol. An automated build-evaluate-iterate loop supporting:

  • Standard Mode: Isolated development in background Git worktrees across a dual-session architecture.
  • Yolo Mode: Immediate inline development branching directly inside the active workspace and terminal.
  • Self-Customization Startup: Auto-configures compilation commands, risk levels, and protected files during first-run initialization.
  • CI/CD Remote Feedback: Polls remote PR checks (e.g., GitHub Actions), parses failure logs, and patches code autonomously.

• delegate: Multi-Model Orchestration & Delegation. Turns Claude Code into an orchestrator that routes each task to the right model — "right model for the right job" — and reports results back. Supporting:

  • Self-Discovery Startup: Detects the model CLIs (codex, gemini, ollama, llm, ...) and Claude models available on your machine, then builds a routing config calibrated to your plans and priorities — not a hardcoded roster.
  • Opt-in Subagents: Ships wrapper subagents that delegate to external models (GPT/Codex, Gemini, OpenRouter, local Ollama, etc.) and only fire on explicit request — safe for governed/brownfield repos.
  • Hardened Delegation: Verified CLI invocations with the stdin-hang, non-TTY stdout, versioned-model-pinning, and sandbox-isolation gotchas already solved.
  • Inspired by @theo (t3.gg)'s posts on model-tiering for agent orchestration and using Codex as a delegation fallback.