Skip to content

Design Intelligence Phase 5 — Marketplace + intelligence #92

Description

@whatiskadudoing

Scope

The polish phase that turns AI Atelie into the design-skill IDE rather than just the place where users author skills. Discoverability + AI-driven assistance + telemetry that sharpens the catalog over time.

Surfaces to build

  • skills.sh embed — browse Vercel's directory inline; one-click "Install into current project." Allowlist via URL pattern (only skills.sh).
  • VoltAgent gallery embed — same shape for VoltAgent's awesome-claude-design's 68 brand-cloned DESIGN.md files. "Use this DESIGN.md" → cloned into the project.
  • AI-authored DESIGN.md — chat-driven: "make me a DESIGN.md for a fintech dashboard, brutalist, dark mode" → agent emits a draft, user iterates inline. Builds on Phase 2's authoring surface.
  • Reverse-engineer scanner — point at the project's existing files; scanner reads tokens, fonts, components from CSS/JS; clusters them; proposes a DESIGN.md draft for review before writing. (The `design-md-author` skill from Phase 1 has this flow documented but un-implemented; Phase 5 actually builds it.)
  • Telemetry — opt-in local logging:
    • which skills were active per prompt
    • DESIGN.md hash on each turn
    • output token volume
    • thumbs up / down on agent output
      Surfaces "skills that work well together," "DESIGN.mds with similar usage," and "your output quality with skill A is +X% vs without" as local stats.
  • Publish flow — eventually, push a user-authored skill to skills.sh / a community gallery (gated behind opt-in OAuth).

Wiring leveraged from earlier phases

  • Phase 2's authoring surface (the agent uses the same editor pane to draft AI-authored DESIGN.md).
  • Phase 3's library (telemetry hangs on the per-user, not per-project, level).
  • Phase 4's import/export flows (skills.sh + VoltAgent embeds use those flows under the hood).

Acceptance criteria

  • User can browse skills.sh from inside AI Atelie and install a skill into the current project
  • User can pick from VoltAgent's gallery and clone a DESIGN.md into the current project
  • AI-authored DESIGN.md flow: chat with the agent, get a complete spec back, save with one click
  • Reverse-engineer scanner: point at an existing project, get a draft DESIGN.md proposal
  • Telemetry runs locally by default (no PII off-device); opt-in to share aggregates

Notes on stability and trust

  • Marketplace install requires user confirmation per skill (no silent installs).
  • AI-authored DESIGN.md drafts always show a diff before writing.
  • Telemetry data lives in `~/.atelie/telemetry/` as JSONL; user can inspect / wipe at any time.

Estimated effort

Open-ended — this phase is more like a backlog of shippable improvements than a fixed-scope sprint. Each surface (skills.sh embed, VoltAgent embed, AI-authored, reverse-engineer, telemetry) is independently shippable.

🤖 Filed via Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions