Ux

UX Thinking Patterns (scored, from skills-db)

Load from ~/.agents/skills-db/thinking/patterns/{name}/SKILL.md when reasoning depth needed:

Hub Shortcuts

UX Pipeline

Mode Parsing

Treat the user's message (including any /ux ... text) as the input.

• If it contains full or interactive: ask a small set of clarifying questions first, then run step-by-step.

• If it contains step N (N=1..7): run only that step and update artifacts.

• If it contains continue: resume the most recent run folder under .lev/ux/.

• Otherwise: AUTO mode. Run all 7 steps end-to-end and produce wireframes.

Output Location

Create a run folder:

• RUN_DIR=.lev/ux/{YYYYMMDD-HHMMSS}-{slug}/

Write artifacts below. If resuming, reuse the existing RUN_DIR.

Artifacts (expected):

• problem_spec.yaml

• routed_skills.json

• jobs.graph.json

• task_graph.json

• ia_schema.json

• interaction_fsm.json

• components.md

• wireframes.md

• summary.md

Phase -1: Setup (Deterministic)

1. Create run dir and establish shell vars:

TS="$(date +%Y%m%d-%H%M%S)"
SLUG="$(printf '%s' "$USER_MSG" | tr '[:upper:]' '[:lower:]' | rg -o "[a-z0-9]+" | head -n 6 | paste -sd- -)"
RUN_DIR=".lev/ux/${TS}-${SLUG}"
mkdir -p "$RUN_DIR"

2. Persist the raw request for traceability:

printf '%s\n' "$USER_MSG" > "$RUN_DIR/request.txt"

Phase 0: Prior Art (Fast)

1. If bd exists and .beads/ exists:

• bd list --status=open | rg -n "ux|design|wireframe|ia|flow" -i || true

2. Scan local UX artifacts:

• ls -t .lev/ux 2>/dev/null | head -n 10 || true

3. Scan UI code quickly (keep it cheap):

• rg -n "struct\\s+.*View\\b|SwiftUI|Navigation(Stack|SplitView)|TabView" apps packages extensions 2>/dev/null | head -n 50 || true

If you find relevant prior art, cite paths in summary.md and either extend it or explicitly supersede it.

Phase 0.5: Agentic UX Patterns (lev-ref)

Apply lev-ref patterns to all UX outputs (especially summary.md and any routed-skill handoff).

This phase doubles as the agentic-ux hub route. Load directly when asked about CLI design, progressive disclosure, or prompt steering — no need to run the full 7-step pipeline.

Key patterns:

• CLI-as-Prompt: Every CLI surface is a prompt-steering opportunity
• Progressive Disclosure: Metadata first, body on trigger, references on demand
• Every Surface a Steering Attempt: Follow-ups, related skills, confidence inline
• Audio Design: Consider sonic feedback for agent interactions

1. Load lev-ref patterns:

sed -n '1,220p' docs/_inbox/lev-ref/SKILL.md

2. Apply inline conventions (CLI-as-Prompt):

• Add confidence lines like [85% confident] for key recommendations.

• Add → Next: and → Related: follow-ups.

• Add a short 💡 Tip: in summary.md if there's a non-obvious UX lesson.

Phase 0.6: Skills-DB Enrichment

Route to specialized skills using the Hub Decision Tree above.

1. Check hub_routes (frontmatter) for trigger match against request keywords.

   • If match -> load that skill, optionally resume pipeline at Step 3+.

2. If no hub_route match, query skills-db dynamically:

# Design-ux category first (high-signal, 12 skills)
rg -l -i "{keywords}" ~/.agents/skills-db/design-ux/*/SKILL.md

# Broader search only if design-ux returns empty
lev-skill resolve "{keywords}" --json 2>/dev/null | jq '.results[:5]'

3. If curated scores available, prefer higher-scored patterns:

grep -i "{keyword}" ~/.agents/skills-db/thinking/11-ui-ux/task.csv \
  | sort -t, -k8 -rn | head -3

4. Emit routed_skills.json (keep existing schema for compatibility):

{ "request_terms": [...], "candidates": [...], "selected": [...] }

5. Present discovered skills:

   • 1 clear match -> propose loading it
   • 2-3 matches -> ranked list, let user choose
   • 0 matches -> continue with built-in UX pipeline

6. Continue pipeline regardless — routing enriches, never blocks.

Step 1: Problem Framing

Write problem_spec.yaml: