Eight AI personas and a phase gate for Snowflake code

What it does

CocoPlus is a plugin for the Snowflake Cortex Code CLI (coco) that wraps data engineering projects in a rigid, phase-gated lifecycle. It delegates work across eight domain-specific AI personas—such as $de for Data Engineer and $cdo for Chief Data Officer—to move projects through requirements, planning, building, testing, and shipping. Execution happens inside isolated git worktrees using only native Coco constructs: Skills, Agents, Hooks, and an AGENTS.md file.

The interesting bit

Rather than relying on a single generalist assistant, CocoPlus runs specialist personas in parallel with checkpoint-validated delivery at each step, and a multi-layer safety gate stands between the agents and your production Snowflake objects. It also tracks tokens, decisions, and quality findings across sessions, accumulating a persistent pattern library that acts as institutional memory.

Key highlights

• Built entirely from native Coco constructs—no external services or extra infrastructure.
• Eight specialist personas cover the full data domain stack, from steward to BI analyst.
• Isolated git worktrees keep parallel agent executions from colliding.
• Multi-layer safety gates aim to prevent accidental modification of production Snowflake objects.
• Tracks decisions, tokens, and quality findings across sessions for auditability and cost visibility.

Caveats

• Tightly coupled to the Snowflake Cortex Code CLI (coco) ecosystem; useless without it.
• The README claims thirty-five features but does not list them, pointing instead to the external docs site at cocoplus.dev.
• Explicitly not an official Snowflake offering, and carries a standard open-source “as is” warranty disclaimer.

Verdict

A compelling addition if you’re already inside the Snowflake Cortex Code CLI and want AI-assisted data engineering to behave more like a governed assembly line than a freestyle script session. If you aren’t using coco, this is just a curiosity.