# SpecOps

A structured planning workflow for AI coding agents that inserts an understand, spec, implement and complete cycle before code is written, keeping requirements and task files in git.

Invoked with one command, SpecOps has the agent first understand the codebase and context, then write a requirements document using EARS-notation acceptance criteria, a design document weighing trade-offs such as JWT against sessions, and an ordered task list with dependencies and effort estimates, before implementing each task against its own written acceptance criteria and a final verified-completion check.

It works natively with Claude Code, Cursor, OpenAI Codex, GitHub Copilot and Google Antigravity, installable as a Claude Code plugin or through a one-line script on any platform. The project targets seven vertical templates, including backend, frontend, infrastructure, data pipelines and library or SDK work, adds a local memory layer that loads prior decisions automatically each session, a configurable team-review gate before implementation starts, and automated drift checks that flag when a spec and its implementation have diverged.

Website: https://github.com/sanmak/specops