Agent Workflow Blueprint

Prompt library + skills + acceptance contracts

Control agent work with evidence, not optimism.

A research-informed workflow starter for teams building with AI agents. It turns human-AI interaction, AI-risk, accessibility, security, and prompting research into 100 copy-ready prompts, skills, contracts, governance artefacts, evidence gates, and verification scripts.

Created by Carla Goncalves · LinkedIn

Browse 100 assets Start here
  • 40prompts
  • 30skills
  • 30contracts
  • 100copy-ready assets

Template product layer

Find the right prompt, skill, or contract.

Search the manifest, filter by type, then open one focused view with an introduction, copy-ready prompt code, and linked references. The static 100-asset index and plain download links remain crawlable without JavaScript.

Loading asset catalogue…

Build an agent or project

Build a governed agent or project from this library.

Launch a browser-local console that captures governance inputs, previews the files being assembled, lets teams remove optional files, applies the human-AI quality standard, and downloads a ZIP starter without using a backend.

Open builder Review proof rails

$ open build-project/

> choose project type

> generate governance pack

> apply quality standard

> download governed starter ZIP

Starter packs

Use bundles, not isolated prompts.

Each pack combines a prompt, a skill, and a contract for one failure mode.

Codex control

Task control pack

Master prompt + scoping skill + completion honesty contract.

Open pack

UI

UI truth pack

Rendered UI audit + screenshot review + visual quality contract.

Open pack

Accessibility

Accessibility remediation pack

Semantic remediation + inspection skill + acceptance contract.

Open pack

Release

Release gate pack

Final merge prompt + packaging skill + release evidence contract.

Open pack

Operating system

The workflow separates doing, reviewing, and claiming.

  1. 1

    Scope

    Define intent, risk, affected files, and evidence gaps.

    Open scoping packet
  2. 2

    Implement

    Make the smallest correct change inside the approved boundary.

    Open implementation prompt
  3. 3

    Review

    Reject scope drift, shallow tests, broken UI, or unsupported claims.

    Open review prompt
  4. 4

    Specialist review

    Use accessibility, security, testing, documentation, or release gates when relevant.

    Open specialist library
  5. 5

    Verify

    Run exact commands and disclose skipped or unavailable checks.

    Open verification script
  6. 6

    Claim only what is proven

    End with verified, partially verified, not verified, or blocked.

    Open rubric

Proof rails

Verification is part of the product.

The scripts check structure, linkage, v3 prompt quality, evaluation coverage, claim discipline, examples, and the 100-asset library.

Checked automatically

  • Asset metadata and dependencies
  • Public claim restrictions
  • Homepage and builder browser flows
  • Automated accessibility rules
  • ZIP extraction and integrity

What has not been verified

  • Universal workflow effectiveness
  • WCAG conformance or assistive-technology coverage
  • Security posture of generated or deployed systems
  • Performance under production workloads
  • Legal suitability for a specific adopter

Examples

The library teaches rejection, not just generation.

Weak vs strong output

See how fake completion and overclaiming get rejected.

Open examples

Research basis

Review the behavioural controls and standards behind the assets.

Open basis

Full catalogue

Open all 100 assets grouped by type and category.

Open catalogue