Turn AI coding into an engineering process.
- 1Classify
gem-orchestratortask routing · context budget
- 2Plan
gem-plannerplan.yaml · acceptance criteria
- 3Implement
gem-implementertests green · reviewed diff
- 4Review
gem-reviewerreview verdict · security audit
- 5Learn
gem-skill-creatorreusable SKILL.md
typical API call on 100K+ token contexts
cheaper than uncached input
DeepSeek V4.1 Flash charges $0.15/M uncached input and $0.003/M cached input. Gem-Team's scoped handoffs keep most calls hitting cache - turning large contexts into sub-penny operations. See the full breakdown
One pipeline, five stages.
Classify
gem-orchestratorReads the task and routes it to the right specialist. Loads only the context the task needs, so sessions stay lean and cheap.
Output: task routing · context budget
Plan
gem-plannerTurns the task into a bounded wave plan: milestones, routing, handoffs, risks, and acceptance criteria. Cuts scope with YAGNI and KISS before any code is written.
Output: plan.yaml · acceptance criteria
Implement
gem-implementerBuilds in TDD cycles - red, green, refactor. Applies SOLID, fail-fast, and least-surprise. Every major decision carries a one-line reason.
Output: tests green · reviewed diff
Review
gem-reviewerIndependent review of plans, code, and decisions. Checks quality, security, and contracts before anything merges.
Output: review verdict · security audit
Learn
gem-skill-creatorPackages what worked into reusable skills, so the team gets better with every session.
Output: reusable SKILL.md
Engineering discipline, enforced.
Works with the tools you already use.
Install the team in one command.
curl -sSL https://aka.ms/apm-unix | sh
apm install mubaidr/gem-team