Human Agency Review

Give people a real choice.

Can someone tell what an action will do, say no without losing unrelated access, and recover if they change their mind? Human Agency Review helps you ask that while coding. Choose one interaction, notice a source clue, consider a clearer alternative, then test the actual interface.

Newer coders can use the seven lessons to make everyday decisions concrete; experienced developers can use them as a focused review before a change ships. No Predict, Eco-Coding, account or AI subscription is needed. Human Agency Review grew from az8t lab's firsthand experience with technology's effects on people and a practical question: how can developers make it easier to understand, choose and recover? It turns that concern into a short review habit, informed by trauma-informed computing and accessibility guidance.

WCAG's limit for flashing content offers one concrete check. A systematic review of cognitive rehabilitation covers attention and memory; these are useful prompts to check whether key instructions stay visible and people can pause or recover without relying on memory. Needs vary. A source clue can prompt a useful check; it cannot tell you what a particular person needs.

Review one interaction

1. Select a complete interaction in a local VS Code file.
2. Run Human Agency: Open Review and choose a topic.
3. Inspect the selected source, or use the illustrated lesson on its own.
4. Read any finding's limit, then verify real behavior in your test tools.

Notice → Understand → Try → Check is the guide's route. Compact mode shows each step together; Guided mode focuses on one at a time. Move freely: there are no scores, timers, compulsory reflections or progress gates.

Each topic includes an example, a possible alternative, a counterexample and checks to adapt. More warnings or confirmations are not automatically better. For instance, a literal Cancel button can submit a form, but its handler may prevent submission; changing its type alone will not stop an in-flight request. Check what your interface actually does before changing it.

Research, responsibilities and scope

Trauma-Informed Computing offers an established framework for safety, trust and enablement. WCAG 2.2 has criteria with specific scopes; W3C cognitive-accessibility guidance adds advice. Together, they give developers practical questions about safety, choice and accessibility to bring into code review.

Accessibility, privacy, consumer-protection and laws addressing fraud or cyber-abuse can matter to professional development. Duties vary by country, service and role. A code pattern is not a legal verdict: check actual behavior and ask the responsible organisation or a qualified adviser when needed. See the local evidence ledger and legal context.

Human review works in any programming language. Interactive teaching examples include Python/Tkinter, C#/WPF, Java/Swing, Kotlin/Android, SwiftUI, HTML and CSS. Bounded source observations cover selected HTML, Vue/Svelte markup, JS/TS JSX/TSX, CSS/SCSS infinite-animation declarations and media autoplay. Exact localized labels for consent and Cancel may be entered for one inspection; they are not retained. Unsupported syntax and runtime behavior stay manual. No findings is not a safety verdict or proof of usability, accessibility or compliance.

Optional Predict connection

Use with Predict to start its normal code-prediction prompt. First predict what the AI code will do. Use your usual AI assistant, accept its suggestion, select that code, then run Predict: Confirm Selected Completion. Return here to test actual interface behavior. This step is optional. The review topic, source, notes and findings stay here; nothing from the guide is pasted into Predict's prediction. Predict compares code, not accessibility or human impact. Press Escape in Predict to leave its prompt. The guide works without it.

Source inspection runs locally, only when requested; it does not execute or change files, scan in the background or upload source. Notes stay in the view until you preview and export a new Markdown file. Existing files are not overwritten. See Human Agency: Open Guide in VS Code for how-to, evidence, privacy and legal context.