원클릭으로
원클릭으로
Branch merging and conflict resolution patterns. Use when merging feature branches, handling conflicts, or cleaning up worktrees. Trigger with "merge branch", "resolve conflicts", "cleanup worktree".
Controls Chrome browser: takes screenshots, clicks buttons, fills forms, downloads images, inspects pages, captures network requests, checks console errors, debugs API issues. Use when: 'screenshot', 'click', 'fill form', 'download image', 'check browser', 'look at screen', 'capture page', 'check for errors', 'debug network', 'API failing', 'console errors'. Provides MCP tool discovery for 70 tabz_* browser automation tools.
Brainstorm ideas and design workflows with a beads expert. Use when: 'help me plan', 'how should I structure this', 'what beads features', 'brainstorm', 'before I create tasks', 'design my workflow'. For ideation BEFORE creating issues.
Docs/changelog checkpoint. Detects whether changes are user-facing and ensures documentation hygiene (CHANGELOG + lessons-learned) is handled or explicitly bypassed. Use when: 'docs check', 'changelog', 'release notes', 'did we update docs?'.
Multi-model conductor - spawns Haiku/Sonnet/Opus/Codex based on task complexity
AI-assisted backlog grooming: prioritize, add dependencies, assign quality gates, draft worker prompts
| name | code-review |
| description | Review code changes with parallel detection agents and smart fixing |
| user-invocable | true |
Language-agnostic code review with dynamic agent planning. Adapts to any language, framework, and project size.
/code-review # Interactive — asks what to review
/code-review --full # Review entire codebase
/code-review all # Same as --full
/code-review --files src/api tests/ # Review specific paths
/code-review <issue-id> # Review changes for a beads issue
/code-review --quick # Fast mode: lint/build check only
SELECT → DISCOVER → SCOPE → PLAN → SCAN (parallel) → AGGREGATE → FIX (if needed) → REPORT
If no flags given, asks what to review (diff, full codebase, specific paths, beads issue, or quick check).
Auto-detects languages, frameworks, linters, and build tools by scanning the project.
Determines what to review based on invocation mode:
git diff HEAD)--full: entire codebase (prioritizes recently changed files for large projects)--files: specific paths<issue-id>: changes for a beads issue (MCP show or bd show)Dynamically decides how many scanner agents (2-8) and what each focuses on, based on:
Launches all planned agents in parallel:
Merges findings from all agents, deduplicates, filters by confidence (>= 80%).
Only if issues found: launches fixer (Opus) with the project's build/lint commands for verification.
Clear pass/fail report with categorized findings.
| Agent | Purpose | Model |
|---|---|---|
| scanner | Generic scanner, invoked N times with different focus prompts | Sonnet |
| claude-md-scan | CLAUDE.md compliance check | Haiku |
| fixer | Apply fixes for high-confidence issues | Opus |
Works with any language. Built-in pattern libraries for:
For unlisted languages, the orchestrator infers patterns from the project structure and CLAUDE.md.
| Scenario | Agents Used | Cost |
|---|---|---|
| Clean code (small diff) | 2-3 Sonnet | $ |
| Clean code (large/full) | 5-8 Sonnet | $$ |
| Issues found | N Sonnet + 1 Opus | $$$ (only when needed) |
| Quick mode | 0 agents | Free |
| Score | Action |
|---|---|
| 0-79 | Skip |
| 80-89 | Flag + TODO comment |
| 90-100 | Auto-fix |
Prefer MCP tools when available, fall back to CLI:
# Get issue context (MCP preferred)
mcp__beads__show(issue_id="<issue-id>")
# CLI fallback: bd show <issue-id> --json
/code-review <issue-id>
# If passed (MCP preferred)
mcp__beads__update(issue_id="<issue-id>", status="reviewed")
# CLI fallback: bd update <issue-id> --status=reviewed
--quick for trivial changes (docs, config)--full for comprehensive codebase review