Patterns, heuristics, and trust systems for AI products.
We don't design interfaces. We design relationships between users and agents.
AUX is a named discipline. This is its public layer — the codified, inspectable, forkable version of what most teams only talk about.
If you ship AI products, you are no longer designing screens. You are designing an ongoing relationship between a user and an agent that has memory, initiative, and judgment. Classical UX was built for tools. AUX is built for relationships.
| Entry point | What it is | Use when |
|---|---|---|
| aux-frameworks | Methodology as code. Audits, taxonomies, scorecards as YAML. | You want the standard. |
| aux-audit | GitHub Action auxfirst/trustkit@v0.2 — scores an agent spec against the 10 heuristics. |
You want a grade, not an opinion. |
| trust-gap-classifier | A taxonomy + detector for how agent trust breaks. | You want shared language for failures. |
| agent-memory-policy | Generate memory governance policy + system-prompt snippets from a config. | Product, compliance, and devs need to agree. |
| agent-ux-teardowns | Structured teardowns of shipped AI products. | You want receipts, not vibes. |
- Definition repos — the "Wikipedia pages" for the category: patterns, heuristics, trust stages.
- Executable repos — CLIs and scanners that turn the ideas into diagnostics.
- Reference implementations — one pattern, runnable, clean, minimal.
- Random experiments, SaaS clones, tutorials, or think-pieces dressed as code.
- Opinions without benchmarks.
- Frameworks without schemas.
One technical blog post with runnable code + benchmark numbers will out-earn ten soft think-pieces.
This is the proof layer for the category — every claim ships with a schema, a CLI, or a reproducible scenario.
Every tool ships as a CLI. Every framework ships as YAML/JSON. Every claim ships with a reproducible scenario.
- uses: auxfirst/trustkit@v0.2
with:
spec: ./agent-spec.yaml
fail-on: highFrom a clone, without npm:
npm --prefix packages/aux-audit ci && npm --prefix packages/aux-audit run build
node packages/aux-audit/dist/cli.js run ./agent-spec.yamlThe spec format (agent-spec.schema.yaml) and rule set (aux-heuristics.yaml) drive the CLI. See packages/aux-audit.
Use the 10 AUX Heuristics and the Trust Architecture to diagnose, plan, and review. Both are editable YAML — fork, extend, submit a PR.
Teardowns are structured (analysis.md + score.json) so they can be compared, aggregated, and cited. Start with agent-ux-teardowns.
See the visual overview: visuals/storyboard.html.
Internal (AUX Studio) Public (GitHub)
───────────────────────────── ─────────────────────
Framework engine ───▶ aux-frameworks (schemas)
Audit engine ───▶ aux-audit (CLI)
Trust taxonomy ───▶ trust-gap-classifier
Memory policy generator ───▶ agent-memory-policy
Teardown templates ───▶ agent-ux-teardowns
The internal system is the private execution layer. GitHub is the public standard layer. Both point at the same taxonomy.
Every repo has a CONTRIBUTING.md. The short version: patterns need diagrams, heuristics need examples, benchmarks need a runnable script. No exceptions.
Definition repos: CC BY 4.0. Executable repos: MIT. Teardowns: CC BY-NC 4.0 (attribution, non-commercial).
auxfirst — agentic UX as a named discipline.