A public-good, government-grade AI knowledge utility

Build safer, fairer and more accountable AI-assisted public services.

A comprehensive portal and toolkit for AI skills, agents, metadata standards, human review, code of practice, secure-by-design delivery, protection by default, quantum readiness and inclusive user-centred design.

Explore the toolkit Read the code of practice

Mission

From static guidance to a living public-sector utility

This portal frames AI-assisted delivery as a governed public good: open where possible, protected where necessary, and continuously improved through evidence, service research and community assurance.

100%material AI-assisted work creates evidence
0self-approval for high-risk outputs
H0–H5risk-based human-in-the-loop model
PQCcrypto-agile and quantum-ready by design

Combined body of knowledge

Authoritative guidance, standards and reusable intelligence

The corpus is organised as a knowledge graph that links laws, policies, standards, design guidance, engineering practice, AI governance, metadata, evidence and reusable delivery patterns.

Guidance corpus

Index GOV.UK, GDS, CDDO, departmental, standards, GitHub and AI Engineering Lab sources with authority tiering, freshness checks and applicability metadata.

Knowledge graph

Link services, repositories, datasets, APIs, controls, skills, agents, people, approvals, models, risks and evidence into one navigable public-sector map.

Executable guidance

Translate policy and standards into schemas, conformance packs, tests, scripts, checklists, prompts, skills and agent guardrails.

Toolkit

A complete operating kit for AI-assisted public delivery

Use the filters to explore reusable components. Each asset is designed to be catalogued, versioned, assured and adapted across tools such as Codex, Copilot, Claude Code and CI pipelines.

AI asset metadata standard

Machine-readable schemas for skills, agents, prompts, tools, model-use events, evidence packs, source records and orchestration runs.

Skill registry

Discover approved, draft and deprecated skills with owners, sources, risk ratings, compatibility, evaluation reports and required reviewers.

Agent registry

Catalogue bounded agents by autonomy level, permissions, allowed skills, prohibited actions, model constraints and escalation rules.

Evidence pack builder

Generate source consultation records, AI-use declarations, validation results, human acknowledgements and approval logs.

DEIA impact screen

Assess equality, accessibility, assisted digital, digital inclusion, protected characteristic and vulnerable user impacts before reliance.

Quantum readiness assessor

Inventory cryptography, identify long-life data, assess harvest-now-decrypt-later exposure and plan crypto-agile migration paths.

Assurance

Human-centred controls for trustworthy AI work

Risk-based human-in-the-loop controls make sure AI does not grant authority the user does not possess. Material work requires evidence, review and accountable ownership.

1

Classify

Identify service impact, data sensitivity, equality and accessibility risk, security risk and model/tool exposure.

2

Select

Recommend applicable standards, source guidance, skills, agents, conformance packs and human review roles.

3

Execute

Apply approved tools with least privilege, source citations, deterministic checks, pair/buddy support and logs.

4

Review

Use 4-eyes or 6-eyes review, specialist assurance and no self-approval for material or high-risk outputs.

5

Learn

Capture evidence, incidents, feedback, source drift, model changes and improvement opportunities.

Metadata catalogue

Treat skills, agents and code as assets in their own right

Every reusable AI capability should be discoverable, accountable, interoperable and governed like a public-sector knowledge, information, data or software asset.

Catalogue fields

  • Asset ID
  • Owner
  • Status
  • Version
  • Authority tier
  • Risk level
  • Source provenance
  • Review roles
  • Compatibility
  • Evidence rules
  • Lifecycle
  • Deprecation
asset_type: ai_skill
asset_id: review-accessibility-govuk
status: approved
human_review: H3 specialist
self_approval_allowed: false
source_freshness_required: true
evidence_pack_required: true

UCD, DEIA and accessibility

Designed for real people, varied needs and accountable teams

The portal is built around inclusive journeys for public servants, suppliers, reviewers, specialists and service owners. DEIA is treated as a design requirement and assurance control, not a final checklist.

New or occasional contributor

Plain-language guidance, confidence prompts, buddying, role limits and safe defaults.

Experienced practitioner

Fast routes to standards, code patterns, conformance packs, test scripts and evidence automation.

Specialist reviewer

Source-linked evidence, issue triage, risk decisions and assurance history.

Accountable owner

Risk overview, approval gates, residual risk, maturity dashboards and public transparency records.

Inclusion commitments

Use semantic HTML, keyboard access, visible focus, high colour contrast, responsive layout, reduced-motion support, plain language, progressive disclosure and routes for assisted support. The content is designed to support equality, diversity, inclusion and accessibility obligations, but formal compliance decisions remain with authorised humans.

Code of practice and conduct

Responsible use of AI assistive technology

This code of practice complements behavioural codes of conduct. It defines operational rules for model-backed tools, direct AI use, coding assistants, multi-agent orchestration and AI-generated artefacts.

Remain accountable, disclose material AI assistance, verify outputs, protect data, cite sources, preserve evidence, avoid unsupported compliance claims and escalate uncertainty.

Do not input unauthorised sensitive data, bypass review, self-approve material work, hide AI use, deploy without authority, override failing checks or claim legal/compliance approval without accountable review.

Use one accountable orchestrator, least-privilege tools, explicit skill versions, separation of duties, conflict escalation, audit logs, evidence packs and human approval before high-impact action.

Make guidance usable, testable and accountable.

Start with the corpus, encode metadata, publish skills, bound agents, enforce review, and continuously improve through evidence and user research.

Back to top