AI skills · built at Oraion

Three AI skills I built for design work.

I built these as the only designer at Oraion. Component Audit checked about 15 components before they counted as production-ready. I used Opinion and Deep Research with the PM on product and research decisions.

The skills.

Component Audit

Component Audit flow A flowchart of the component audit skill: a selected Figma component has its layers renamed, then a variable map is built from the Base Elements library. Properties that match a token are bound automatically; those that do not become proposals for new tokens in Base Elements. Both paths rejoin at a design review panel covering UX, UI and copy, and the run ends by appending a dated entry to the audit log. ONLY SOURCE MATCH NO MATCH PROPOSE Select component AUTO Rename layers defaults + casing drift AUTO Build variable map value → token lookup LIB Base Elements every variable lives here Exact token for this value? AUTO Bind the variable h/v spacing aware DESIGNER Propose a token named from neighbours Design review panel UX · UI · COPY scored, tiered, with what's working well Append dated log entry LEGEND AUTOMATED STEP DECISION LIBRARY NEEDS A DESIGNER REVIEW · FEEDS LIBRARY

It renames layers, builds a value-to-token map from Base Elements, binds every property that matches, and turns the ones that don't into proposed tokens. Then a three-part review covering UX, UI and copy, and a dated entry written to the log.

Opinion

Opinion flow A flowchart of the opinion skill: a request is clarified for research depth and scope, then branches. A quick search stays in the conversation, while the full pipeline hands off to the deep research skill and returns with a notebook. Both paths rejoin to produce a SWOT analysis, a synthesising overview, and finally a direct recommendation that names the trade-off which would change it. FULL PIPELINE QUICK RETURNS "What do you think?" ASKS Clarify depth + scope a check, not an interrogation How deep does the research go? DELEGATES Run deep-research EVALS: 40% → 100% ONCE WIRED notebook + optional Notion AUTO Quick search 5–8 sources, no artifacts AUTO SWOT 2–4 cited bullets, no sugarcoating AUTO Overview synthesis, not a recap A position, plus its caveat LEGEND AUTOMATED STEP DECISION ANOTHER SKILL, CALLED

Structured decision-making: research, SWOT, synthesis, and a direct recommendation.

Deep Research

Deep Research flow A flowchart of the deep research skill: a request is clarified for angle and source preferences before any searching, then four to six searches run with a domain blocklist appended to every query. Sources are added to a NotebookLM notebook; any host that fails silently is added to the blocklist for next time. The notebook is the durable artifact, logging to Notion is optional, and follow-up questions are answered from the sources themselves with citations. APPENDED FAILED ADD HOST Research request ASKS Clarify angle + sources two questions, before searching AUTO Run 4–6 searches target 15–20 sources LIB Domain blocklist hosts that fail silently AUTO Add each source url + youtube auto-detected SKIPPED Blocklist the host no repair attempted NotebookLM notebook THE DURABLE ARTIFACT audio overview, mind maps, flashcards in the Studio panel OPTIONAL Log to the Notion hub asked, never assumed Answer, cited from sources LEGEND AUTOMATED STEP LIBRARY OPTIONAL THE ARTIFACT

Automated topic research: search, source-quality filtering, logging.

Scaling this across a team is the same move each time: build the skill, write the evals so you know whether it works, and hand over the ones built to be handed over.

Every skill here started as a fix for my own work. The best ones made my team's work better too. If that's the designer you're looking for, let's talk.

Let’s talk