Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
carlonicora avatar

Architect

  • 3 repo stars
  • Updated February 14, 2026
  • carlonicora/cc-architect

Development architect for claude code using OpenSpec and Beads

About

architect is a Claude Code skill in the AI & Agent Building category. Development architect for claude code using OpenSpec and Beads

  • architect
  • AI & Agent Building
  • AI-coding skill

Architect by the numbers

  • Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add carlonicora/cc-architect
/plugin install architect@cc-architect

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
repo stars3
Last updatedFebruary 14, 2026
Repositorycarlonicora/cc-architect

What it does

Development architect for claude code using OpenSpec and Beads

README.md

Architect for Claude Code

Plan → Spec → Beads → Execute. Persistent task management that survives context compaction and session boundaries.

The Problem

Claude Code is powerful, but without structure it can:

  • Start coding before understanding the full picture
  • Lose track of what's done when context gets long
  • Say "done" when tests are still failing
  • Hallucinate on large features that exceed context

The Solution: Break work into self-contained beads with full implementation details.

/architect:openspec "Add user authentication with JWT"
/architect:review                   # Interactive: validate specs before implementation
/architect:beads                    # Interactive: select from available specs
/architect:implement                # Interactive: select epic and execution mode

Install

cc-architect Plugin

# Add to marketplace
/plugin marketplace add carlonicora/cc-architect

# Install the architect plugin
/plugin install architect@cc-architect

Dependencies

Tool Install Required For
OpenSpec See OpenSpec installation /architect:openspec, /architect:review
Beads brew tap steveyegge/beads && brew install bd /architect:beads, /architect:implement

Commands

Command Purpose Interactive
/architect:openspec [plan|task] Create OpenSpec proposal Optional input
/architect:review [spec-path] Validate and edit specs Yes - sprint-meeting style
/architect:beads [spec-path] Convert spec(s) to beads Yes - lists available specs
/architect:implement [--label|--mode] Execute beads Yes - select epic and mode
/architect:cancel Stop active execution No

Workflow

PLANNING                              EXECUTION
┌─────────────────────────────────┐   ┌─────────────────────────────────┐
│ 1. /architect:openspec <task>   │──▶│ 3. /architect:beads             │
│ 2. /architect:review            │   │ 4. /architect:implement         │
└─────────────────────────────────┘   └─────────────────────────────────┘

Stage 1: Create Proposal

/architect:openspec "Add billing integration with Stripe"

Creates an OpenSpec proposal at openspec/changes/<id>/ with:

  • proposal.md - Overview and scope
  • design.md - Reference implementation code
  • tasks.md - Task breakdown with exit criteria

Stage 2: Review Spec

# Interactive mode (recommended):
/architect:review
# Lists available OpenSpec changes and walks through validation

# Direct mode:
/architect:review openspec/changes/billing/

Sprint-meeting-style validation that walks through your spec section by section:

  1. Proposal Review — Why, What Changes, Impact, Success Criteria, Out of Scope
  2. Requirements Review — For each requirement:
    • Business Logic: Is the requirement correct?
    • Architecture: Is the technical approach sound?
    • Development: Are implementation details correct?
  3. Design Review — Reference Implementation, Migration Patterns, Risk Analysis
  4. Tasks Review — Phases, Exit Criteria

Answer "Yes" to continue, "No" to discuss and edit. Changes are applied immediately with cascading updates to related files.

Why review? Editing specs is cheap; debugging bad beads is expensive. Catch issues before code is written.

Stage 3: Import to Beads

# Interactive mode (recommended):
/architect:beads
# Lists available OpenSpec changes and asks which to convert

# Direct mode:
/architect:beads openspec/changes/billing/

# Multiple specs (from auto-decomposed proposal):
/architect:beads openspec/changes/billing-backend/ openspec/changes/billing-frontend/

Auto-decomposition: Beads >200 lines are automatically split—parent marked decomposed, child sub-beads created.

Stage 4: Execute

# Interactive mode (recommended):
/architect:implement
# Lists available epics, then asks execution mode: Swarm, Loop, or Auto Loop

# With specific label:
/architect:implement --label openspec:billing

# With specific mode:
/architect:implement --mode swarm --workers 5
/architect:implement --mode loop
/architect:implement --mode auto

Cancel execution:

/architect:cancel

Execution Modes

Mode Description Best For
Swarm Parallel workers (default: 3) Independent tasks, maximum speed
Loop Sequential with pauses Interdependent tasks, debugging
Auto Loop Sequential, no pauses Hands-off sequential execution

Choose Loop/Auto Loop when:

  • Tasks depend heavily on each other
  • You want to review each step (Loop mode)
  • Debugging issues

Choose Swarm when:

  • Many independent tasks
  • Want maximum speed
  • Large refactoring with parallel phases

Progress Visualization

Press ctrl+t at any time during execution to see:

  • Active workers (swarm)
  • Tasks in progress
  • Completed tasks
  • Pending tasks

What are Beads?

Atomic, self-contained task units in a local graph database. Unlike session-scoped plans, beads persist across sessions.

Component Purpose
Requirements Full requirements copied verbatim
Reference Implementation 20-400+ lines based on complexity
Migration Pattern Exact before/after for file edits
Exit Criteria Specific verification commands
Dependencies depends_on/blocks relationships

Key insight: When context compacts, bd ready always shows what's next. Each bead has everything needed—no reading other files.

Essential bd Commands

Command Action
bd ready List tasks with no blockers
bd show <id> Show full task details
bd update <id> --status in_progress Start working on task
bd close <id> --reason "Done" Complete task
bd list --status in_progress Find current work

The Self-Contained Bead Rule

Each bead must be implementable with ONLY its description.

Bad Good
"See design.md" FULL code (50-200+ lines) in bead
"Run tests" npm test -- stripe-price
"Update entity" File + line numbers + before/after

Best Practices

  1. Review specs before beads — Use /architect:review to validate business logic, architecture, and implementation details
  2. Exit criteria are non-negotiable — Not "tests pass" but exact commands: npm test -- auth
  3. Use Loop mode for complex tasks — Prevents quality degradation
  4. Use Swarm for independent tasks — Maximizes parallelism
  5. Review beadsbd list -l "openspec:<name>" then bd show <id> to verify code snippets

Stealth Mode

For brownfield development, keep tracking files out of git:

bd init --stealth    # Adds .beads/ to .gitignore

Context Recovery

If you lose track during execution:

bd ready                        # See what's next
bd list --status in_progress    # Find current work
bd show <id>                    # Full task details

Resources

License

MIT

Related skills

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.