V1 Roadmap: Claude-First Local Skill Workbench
Purpose
V1 is the Claude-focused release of Skill Workshop RAG. It should remain a local desktop tool for managing Claude Code skills, diagnosing setup readiness, asking RAG-backed questions, and safely using either the Anthropic API provider or local Claude CLI subscription mode.
V1 should not implement the provider-agnostic V2 registry. The right V1 move is to polish the current app, deepen Claude Code ecosystem awareness, and add guided learning flows for creating better skills.
Navigation
Current Product Shape
The app is already more than a basic RAG chat:
- Settings can configure workspace paths, API mode, local Claude CLI mode, profile selection, and runtime-applied provider settings.
- Setup Doctor gives deterministic readiness checks for workspace, RAG index, provider mode, Claude CLI, and login helper state.
- The RAG index persists sanitized metadata in
.next/cache and reports ready, stale, missing, rebuilding, or failed.
- Skill create/edit/delete flows validate local skill files and mark the index stale after writes.
- Chat shows provider/index readiness, streams answers, retries failed assistant responses, and links citations back to skills.
- Export can include sanitized diagnostics.
V1 roadmap work should make those capabilities easier to trust and easier to use before broadening beyond Claude.
Phase Order
Phase 1: Release Polish
Make the current product shippable for local users.
Acceptance criteria:
- First-run path, provider setup, and index state are easy to understand from Settings.
- Chat blocks or warns clearly when provider or index readiness is not good enough.
- Setup Doctor and diagnostics avoid secrets, full home paths, raw provider output, and account identifiers.
- Browser smoke checks cover Settings, Chat, Skills, and Export.
Phase 2: Skill Lifecycle
Turn the app from a skill editor into a skill lifecycle manager.
Acceptance criteria:
- Users can create skills from templates.
- Users can import skills from local folder, archive, or GitHub URL only after preview and validation.
- Skill Quality Doctor checks official-style frontmatter, description usefulness, supporting-file references, token weight, dynamic command usage, and unsafe patterns.
- Skill changes produce clear index freshness state and safe rollback guidance.
Phase 3: Claude Code Cockpit
Add read-only awareness of the wider Claude Code project surface.
Acceptance criteria:
- The app can inventory project-level
.claude/skills, .claude/commands, .claude/agents, .mcp.json, settings files, hooks, and plugin-style folders.
- The UI explains what exists, what is missing, and what needs restart or reload, without writing config automatically.
- Setup Doctor can include Claude ecosystem checks without leaking local account or config details.
Phase 4: Guided Learning
Use Claude for Education style ideas to coach users through skill creation.
Acceptance criteria:
- Guided Skill Builder asks users to reason through purpose, triggers, inputs, boundaries, examples, and success criteria.
- The builder gives rubric feedback instead of simply generating a finished skill.
- Templates and previews help users learn what good skills look like.
- Generated or assisted content remains editable and validated before writing.
V1 Release Definition
V1 is release-ready when:
- Local setup can be diagnosed without reading raw env files.
- A new user can select a workspace, rebuild the index, test provider auth, create or import a skill, ask a cited question, and export diagnostics.
- Claude-specific integrations stay local-only where they touch the device.
- Docs clearly distinguish current features from planned enhancements.
- V2 provider-agnostic work remains documented separately under
docs/v2-roadmap/.
Next Implementation Target
Start with docs/v1-spec/iteration-01-release-polish.md. It has the highest leverage because it validates the current app before adding larger Claude ecosystem surfaces.