A Claude Code plugin for planning and executing complex features with reliable verification.
Break large features into smaller pieces, then execute them one by one - with built-in verification that ensures each piece is actually complete before moving on.
Uses Beads as the source of truth for plans and task state. Pull-based execution means you work on one task at a time, with quality gates enforced by hooks.
Prerequisites: Bun, Beads, Git, GitHub CLI
Optional:
- Perles - Visual tracking UI for Beads
- Azure CLI with DevOps extension (
az extension add --name azure-devops) - For Azure DevOps integration (PRs, work item sync) ANTHROPIC_API_KEYenvironment variable - For AI-powered test quality verification
claude plugin marketplace add NotMyself/claude-dotnet-marketplace
claude plugin install planStart planning a new feature. Claude will ask questions, help you think through edge cases, and create a structured plan stored in a Beads epic.
plan:new
Break your plan into executable features. Creates Beads tasks with full prompts and supporting files.
plan:optimize <epic-id>
> plan:new
Claude: What feature are you building?
You: I want to add a health check endpoint
Claude: [asks clarifying questions, creates plan, stores in Beads epic]
Epic created: health-check-abc
> plan:optimize health-check-abc
Claude: [creates Beads tasks with prompts]
> bd ready
> bd update <task-id> --status=in_progress
[work, then close when done]
For complex features with multiple sub-features:
> plan:new --master
Claude: What initiative are you building?
You: Real-time chat with rooms, users, and message history
Claude: [creates Epic with child Features - brief descriptions only]
Epic created: chat-app-xyz
Features: F001 Project Setup, F002 Auth, F003 Rooms, ...
> plan:new chat-app-xyz.f001
Claude: [detailed planning for F001, updates Feature description]
> plan:optimize chat-app-xyz.f001
Claude: [evaluates for research needs, then creates Tasks under F001]
> bd ready
> bd update <task-id> --status=in_progress
[work on tasks, close when done, then plan:new the next feature]
Research evaluation: When optimizing a feature, Claude scans for uncertainty markers (TBD, TODO, unclear, investigate, etc.). If found, it suggests creating research tasks first. Research tasks are immediately workable and bypass workflow checks.
Workflow is enforced - you cannot skip steps. Trying to implement a Feature directly will be blocked until you run plan:optimize.
Create .planconfig in your project root to configure verification commands:
build_command: "npm run build"
test_command: "npm test"
lint_command: "eslint ."
format_command: "prettier --check ."
static_analysis_command: "sonar-scanner"All commands are optional. If not configured, those verification steps are skipped.
Beads tracks all state. Use bd ready to see available work, bd show <task-id> to review any task.
MIT