Show HN: Agentic Code – Make AI coding tools follow real workflows (github.com)

Agentic Code is an open-source framework (npx github:shinpr/agentic-code) that wraps AI coding tools in built-in, professional workflows so an LLM follows project conventions automatically. Drop AGENTS.md and a .agents/ directory into a repo and the agent starts by reading task-analysis.md, loading language/rule files, generating test skeletons, producing design docs for larger features, and enforcing quality gates and approval points. v0.2 adds “tests-first” behavior; other features include progressive rule loading (only load rules needed for a task), metacognition/error recovery, plan injection to force required steps, and a 1-commit-per-task principle to keep history atomic. It works with any AI that understands AGENTS.md (Codex, Copilot, Cursor, etc.), supports any language (TypeScript optimizations available), and is fully customizable. For AI/ML engineering teams this addresses common pain points — LLMs losing project context, deleting tests, ignoring architecture, or skipping QA — by making agent behavior reproducible, auditable, and enforceable. Technically, tasks define WHAT, workflows define HOW, and rules define quality standards; the agent analyzes repo size to choose direct execution vs structured workflow and blocks progress until tests and gates pass. As an MIT-licensed tool built on the AGENTS.md community spec, it’s designed to standardize agent behavior across tools and scale AI-assisted development toward reliable, team-ready workflows.