UI Ideas

Generate 3 novel UI upgrade concepts for AI coding agents. For each implementable idea, build it as a Claude Code skill immediately.

Important

• Read references/catalog.md FIRST to know all 49 existing concepts. Never duplicate them
• Every idea must pass the smoke test: "Does this solve a real friction point, or is it just aesthetic/gamification?"
• Ideas that fail the smoke test are discarded silently — generate replacements until you have 3 passing ideas
• When an idea is implementable as a Claude Code skill, write the SKILL.md file immediately
• Quality over novelty. A solid, useful idea beats a clever but impractical one
• Take your time reasoning about real friction points before generating ideas

Instructions

Step 1: Load prior art

Read references/catalog.md. Internalize:

• All 49 existing concepts (to avoid duplication)
• The 6 problem categories: visibility, efficiency, trust, decision-support, safety, context-management
• What's already implemented vs. what's still conceptual

Step 2: Identify friction points

Reason about real problems developers face when working with AI coding agents. Think about:

• Where do sessions go wrong?
• What information is missing when making decisions?
• What repetitive tasks could be automated?
• Where does trust break down between user and agent?
• What context is lost between sessions?
• What mistakes keep happening?

Generate 5-6 candidate friction points. Discard any already addressed by the catalog.

Step 3: Generate ideas

For each of the best 3 friction points, design a solution:

Name: Short, memorable name (2-3 words)

Category: One of visibility / efficiency / trust / decision-support / safety / context-management

Problem: 1-2 sentences describing the specific friction point

Solution: 2-3 sentences describing what the tool does

Smoke Test: Answer honestly — "Is this genuinely useful or just aesthetic?" If aesthetic, discard and generate a replacement.

Implementability: One of:

• SKILL — Can be built as a Claude Code skill right now
• HOOK — Can be built as a Claude Code hook
• CLI — Requires changes to the Claude Code binary (not buildable)
• PARTIAL — Core value deliverable as skill, but full vision needs CLI

Step 4: Implement buildable ideas

For each idea rated SKILL or HOOK:

1. Create the skill directory: ~/.claude/skills/[skill-name]/
2. Write SKILL.md following the standard format:
   • YAML frontmatter with name, description (trigger phrases + negative triggers), user_invocable: true
   • ## Important section with non-negotiable rules
   • ## Instructions with numbered steps in imperative voice
   • ## Error Handling with numbered conditions
   • ## Examples with 2-3 concrete invocation examples
3. Verify the skill description is under 1024 characters
4. Ensure trigger phrases don't overlap with existing skills

For CLI ideas, describe what would need to change in the CLI but do not attempt implementation.

Step 5: Update catalog

Append the 3 new concepts to references/catalog.md so future invocations don't duplicate them.

Step 6: Output summary

UI IDEAS — [date]

IDEA 1: [name] ([category])
  Problem: [1 line]
  Solution: [1 line]
  Smoke test: PASS
  Implementability: [SKILL/HOOK/CLI/PARTIAL]
  [If built]: Created ~/.claude/skills/[name]/SKILL.md

IDEA 2: [name] ([category])
  ...

IDEA 3: [name] ([category])
  ...

BUILT: [N] new skills created
CATALOG: Updated with [N] new entries ([total] total concepts)

Error Handling

1. Cannot generate non-duplicate ideas: "The catalog covers [N] concepts across all 6 categories. The remaining gaps are in [categories]. Focusing there." If truly exhausted: "All major friction points are covered. Consider deepening existing skills instead of creating new ones."
2. Skill name conflicts with existing skill: Append a differentiating suffix or choose a different name.
3. Idea fails smoke test: Discard silently and generate a replacement. Do not output failed ideas.

Examples

Example 1: Standard invocation

Input: /ui-ideas Process: Reads catalog, identifies 3 uncovered friction points, generates ideas, builds 2 as skills Output: Summary with 3 ideas, 2 new skill files created, catalog updated

Example 2: Repeated invocation

Input: /ui-ideas (second time) Process: Reads updated catalog (now 52 concepts), finds 3 new friction points in underserved categories Output: 3 different ideas from first run, no duplicates

Example 3: Exhaustion

Input: /ui-ideas (after many runs) Output: "The catalog covers 65+ concepts. Remaining gaps are narrow: [specific areas]. Generated 2 ideas in those gaps and 1 that deepens an existing concept."