Method / evidence before invention
Research once. Build from a clear starting point.
Five stages
What is already done for you.
Source links stay visible because this is AI-assisted research, not formal assurance. The quickest way to trust it is to check the trail yourself.
- 01
Start with the strategy
Each playbook begins with an opportunity named in Northern Ireland’s draft AI strategy. We translate the example into plain English without deciding what should be built.
Output / A bounded opportunity, not a product brief
- 02
Investigate the public evidence
AI-assisted research helps identify and interpret published sources. The playbook records the publisher, coverage, access conditions and relevance so builders can verify the research themselves.
Output / A traceable source register
- 03
Prepare safe working data
Where it is responsible, we author a small non-sensitive dataset shaped by fields and vocabulary visible in those sources. Where a person-shaped stand-in would mislead, we supply a refusal instead.
Output / Synthetic working data—or an explicit stop
- 04
Package the domain context
Every playbook has a repository skill and domain brief. After cloning, a coding agent can load the sources, known unknowns and safety boundaries before discussing possible directions.
Output / A domain build partner in .agents/skills
- 05
Keep judgement with the builder
The playbook offers evidence and constraints, not a recommended application. Builders still need affected people, service owners, professional reviewers and data authorities before operational decisions.
Output / A credible place to start—not permission to deploy
Check the source
Follow the evidence yourself.
The opportunity list comes from Table 2 — potential public-service applications. Every playbook then links the public sources used to interpret its domain.
Read the draft strategy consultation (opens in a new tab)Look inside a skill
The skill is part of the product.
Skills are checked into the repository, reviewed like code and validated against every registered playbook. They follow the open Agent Skills standard, so Claude Code, Codex and other compatible agents can load them.
Inspect an example build-partner skill (opens in a new tab)Ready to inspect the work?