agentic-coding-workflow

What problem does it solve?

Provides a repeatable discipline to prevent unlogged, unauthorized, or unsafe autonomous changes during AI-assisted development by enforcing an append-only permissions audit, verification checkpoints, and pre-deployment gates.

Core Features & Use Cases

• Permissions & Audit Logging: Mandates creating and appending to an AUDIT.log at repository root for every autonomous action and records explicit approvals.
• Verification-Before-Implementation: Prescribes read-first verification sessions, written work plans, and Cory approval checkpoints before touching production-affecting files.
• Multi-Model & Deployment Gates: Defines roles for Claude.ai and Claude Code, pre-Railway-push schema and test checks, and a formal session end protocol to ensure safe pushes.
• Use Case: Run this workflow at the start of Cambium sessions, before Railway pushes, and whenever enabling YOLO/autonomous actions to maintain traceability and reduce deployment risk.

Quick Start

Start a session by appending a SESSION_START entry to AUDIT.log, run a verification-only read of the relevant code, create a work plan with approval checkpoints, and obtain Cory's explicit approval before making any changes.