What this is
Twenty-one structured methods, one orchestrating agent
Each skill is a complete AI prompt encoding an established design ethics method — from Brignull's dark-pattern taxonomy to Fogg's behavior model to Stanford's STF-ET tool chain. Together they make rigorous ethical analysis available inside the tools designers and product managers already use, producing concrete artifacts: stakeholder maps, dark-pattern audits with named statutes, behavioral forecasts with cascade analysis, signed ethical contracts.
Why
Access, not replacement
The discipline is rich and well-researched. What's harder is reaching it on demand. A product team that can't add a full-time ethical-design specialist can still bring 21 research-based methods into a sprint review with a single prompt. The skills don't replace expertise — they integrate it, so that ethical thinking happens as part of shipping rather than after.
What's inside
21 methods
| Skill | What it produces |
|---|---|
| Another Lens | Surfaces designer bias and converts insight to a Design Decision Spec decide |
| Anti-Heroes | Names manipulative design moves with a shared card deck and pairs each with a Hero counter-move forecast |
| Bad Design Canvas | 12-category adversarial audit of a product's potential harms audit |
| Black Mirror Brainstorming | Dystopian misuse scenarios to surface non-obvious risks forecast |
| CIDER | Audits exclusionary assumptions embedded in a design audit |
| Critical Interviewing | Research protocol with non-obvious harms inventory and interview guardrails decide |
| DAH Cards | Six harm categories with manifesto option audit |
| Digital Ethics Compass | Four-category audit (data, manipulation, transparency, automation) with stakeholder map and objective-function risk table audit |
| Ethical Contract | Cross-disciplinary signed commitment with bias audit and red lines align |
| Ethicography | Analyzes team decisions over time for ethical trajectory and 12-month forecast audit |
| Fair Patterns | Dark pattern audit with jurisdiction-specific statutes and vulnerable-population matrix audit |
| Humane Design Guide | Six-sensitivity audit with named mechanisms and exploitation-stack analysis audit |
| Inverted Behavior Model | Behavior forecast with worst-possible-design, convergence check, and 5-stage cascade forecast |
| Motivation Matrix | Maps how a product works on five motivations — achievement, social acceptance, fear, power, incentive — for specific users in specific contexts forecast |
| Normative Design Scheme | Three-lens decision support with Universal Law Test and Triad Conflict Matrix decide |
| Pledge Works | 5-part operationalized pledges with "what we refuse to build" register align |
| Responsible Design Prism | Five-axis ethical posture rating with stakeholder map and mechanism audit audit |
| STF-ET | Stanford's Ethics Toolkit: five chained tools (Explore → Evaluate → Decide) ending in weighed options and named value trade-offs forecast |
| Value Dams and Flows | Maps stakeholder value conflicts with power analysis align |
| Values Levers | Identifies levers given the user's role to shift culture toward ethical design align |
| Worrystorming | Structured worry session that reframes concerns as design values forecast |
Foundation
Every method cites its source.
Built on published research, not invented principles
None of these skills were made up. Each one encodes an established design ethics method — Brignull's dark-pattern taxonomy, Fogg's behavior model, Stanford's STF-ET toolchain, the Center for Humane Technology's sensitivity framework, and others. The discipline is four decades deep. This box makes it reachable at the moment a decision is being made, not after.
First, each skill was compared against a strong baseline: a capable model told the method's name and asked to apply it thoroughly. Two independent judges, Claude Sonnet 4.6 and Gemini 2.5 Pro, scored the pairs. 17 of 21 skills won under both, four won under one judge by a thin margin, and none lost under both.
Beating a baseline doesn't show that a skill does what it promises. Every skill ends with a quality bar: what a good output has to contain. We're turning those bars into automated checks and running each skill many times against them. Eleven skills are covered so far.
The checks found real gaps. One skill's card deck lived only in a reference file, so the model invented cards that don't exist. Another asked teams for a commitment with no deadline and no owner. A third offered example heuristics that got pasted into audits of unrelated products. Fixing five of them:
- Anti-HeroesUsed only real deck cards in 2 of 18 runs → 17 of 18
- DAH CardsWrote the required microcopy rewrite in 13 of 24 audits → 24 of 24
- CIDERClosed with a commitment someone can be held to in 28 of 36 runs → 36 of 36
- Humane DesignWrote heuristics specific to the product in 5 of 18 audits → 18 of 18
- EthicographyNamed the affected populations for 23% of decisions → 84%
The checks also show what a skill adds. Asked to run CIDER without the skill, a capable model invented what the acronym stands for in two of three runs.
Outputs generated with DeepSeek V4 Pro and scored with TypeSafe's Jev, calibrated against hand-checked labels. A fix is kept only if the difference holds up under a significance test.
Automated evaluation can tell you whether a skill follows its own methodology. It can't tell you whether that methodology produces insight a practicing designer would trust. That takes human evaluators — designers, PMs, ethicists — reading blinded pairs and reporting which output they'd actually use.
Compare baseline vs. with-skill outputs for all 21 methods. Export your answers and open a GitHub issue. Negative results are as valuable as positive ones.
Start human review5-minute quick start
Try it right now
You don't need to install anything. Open Claude or Codex and try one of these prompts:
If you're designing a feature
"Here's our [flow or screen]: [describe it]. Run an Anti-Heroes critique — which manipulative moves is it making, where exactly, and what's the Hero version of each?"
This uses the Anti-Heroes method
If you're reviewing existing design
"Review this [product/page/design] for dark patterns or manipulative design elements. List each with the type and potential impact."
This uses the Fair Patterns method
If the team is stuck on a decision
"Our team is debating [decision]. Help me break this deadlock by examining it through three lenses: the intention behind it, its effects on everyone involved, and the duties and rights it touches."
This uses the Normative Design Scheme method
Get started
Two paths in
Through the agent
1- In Claude Code, install the plugin:
/plugin marketplace add abektes/edbx-designskills, then/plugin install edbx@edbx-designskills. Elsewhere, clone the repository or add it to your AI assistant's context - Run
/edbx:helpor point your AI assistant at the repo, and describe your situation (e.g., "I'm designing a notification system and worried about notification fatigue") - The
AGENT.mdorchestrator routes you to the right method — no need to pick one yourself - Follow the agent's guidance to get your ethical analysis artifact
Pick a skill directly
2- Browse the
tutorials/directory or the skill index above - Choose a method that matches your current need (e.g., "Fair Patterns" for dark pattern audits)
- In Claude Code with the plugin installed, run it as
/edbx:<name>(e.g.,/edbx:fair-patterns). Elsewhere, load that skill'sSKILL.mdfile into your AI agent as a system prompt - Paste your situation and get the structured output