Introduction
What Skaftor is, the problem it solves, and how a feature travels from a PRD to production through an approval workflow that agents execute and people govern.
Skaftor is the system of record for AI-native software delivery. You bring a product requirement document (PRD) — or just a sentence — and Skaftor turns it into approved requirements, a technical design, and pull-request-sized work orders. Coding agents build those work orders on managed workstations or in your own editor, and every step is approved by a person, checked by gates, and remembered.
When AI agents write most of the code, the bottleneck moves from typing to coordination, governance, and memory — the Orchestration Gap. Skaftor exists to close it.
How a feature moves through Skaftor
Every feature follows the same journey, shown on its page under "Where this is":
- Requirements — Skaftor reads your PRD, asks about anything the document leaves open, and drafts requirements with acceptance criteria. The Product Manager approves them.
- Design — an agent reads your repository and writes the feature's technical design, with system and user-flow diagrams. The Tech Lead approves it.
- Work orders — the design is cut into work orders, one pull request's worth of work each, mapped to the repository it changes.
- Build — coding agents (Claude Code, Codex, Gemini, DeepSeek, or your editor's agent) implement each work order and open a pull request. CI, an AI review and a clause gate check the diff against the acceptance criteria.
- Gates — deployment gates generated from your code decide whether the build is clear to deploy.
- Deploy — Skaftor deploys to your cloud, or records what your own pipeline deployed. QA runs test cases (optionally with the QA agent) before production.
Behind the journey is a configurable workflow — by default Product Documentation, Requirements Approval, Blueprint Approval, Work Order Creation, Development, Pull Request Review, Deployment Approval, QA Approval, Release PR and Production Release — so the right role signs off at each step.
What makes it different
- People approve, agents execute — approvals stay with your team even when stages run automatically.
- Grounded in your code — designs, work orders and gates are generated from your indexed repositories, not a blank page.
- Bring your own AI key — Skaftor runs on your organisation's Anthropic, OpenAI, Gemini or DeepSeek key, with budgets you set.
- Everything is linked — requirement, design section, work order, pull request and deployment are connected in an engineering memory graph.