Danizee Claude Suite

Unified workflow shortcuts for Claude Code that compound knowledge — each task makes future tasks easier.

Installation

npx danizee-claude-suite init

Options:

npx danizee-claude-suite init --force         # Overwrite existing files
npx danizee-claude-suite init --with-pm       # Include PM workflows
npx danizee-claude-suite init --without-cookbook  # Skip Agent Cookbook
npx danizee-claude-suite init --dry-run       # Preview without writing

Requirements

• Node.js >= 20.0.0
• GitHub CLI (gh) — recommended for PR workflows

How It Works

Every workflow follows the Search → Execute → Verify → Compound pattern:

 1. SEARCH — Check memory + Pi Brain for past solutions
 2. EXECUTE — Run the workflow with checkpoint gates
 3. VERIFY — Cross-method validation (files, tests, git diff, build)
 4. COMPOUND — Store solution + Ralph candidates + RC-A candidates

The second time you solve a similar problem, it's faster because the workflow finds and applies your previous solution.

Quick Reference: /w- Shortcuts

Development

| Shortcut | Description | |----------|-------------| | /w-tdd-swarm | Full TDD + Swarm — plan, test-first, parallel build, review | | /w-plan-tdd-swarm | Interview refines idea → Full TDD Swarm builds it | | /w-swarm | Parallel agents (coder, tester, reviewer) | | /w-autoresearch | Autonomous experiment loop for measurable optimization | | /w-agent-tdd-swarm | Gateless TDD for terminal agents (zero user gates, auto-PR) | | /w-agent-interview-swarm | Interview → spawn gateless terminal agent |

Bug Fixes

| Shortcut | Description | |----------|-------------| | /w-fix | Quick fix — investigate → TDD fix → verify | | /w-debug | Deep debug — hypothesis → diagnose → TDD fix | | /w-hotfix | Critical production fix — isolated branch + security review |

Reviews & Audits

| Shortcut | Description | |----------|-------------| | /w-review | Comprehensive code review (12+ specialized agents) | | /w-security | Security audit — OWASP top 10, auth/authz, data exposure | | /w-perf | Performance audit — N+1 queries, memory, bottlenecks |

Architecture

| Shortcut | Description | |----------|-------------| | /w-architect | Hive-mind architecture — collective intelligence for complex design | | /w-multi-repo | Coordinate changes across repositories |

Session Management

| Shortcut | Description | |----------|-------------| | /w-start | Cold-start session with project context | | /w-end | End session — compound knowledge + commit | | /w-search | Search past solutions in memory |

Compound & Knowledge

| Shortcut | Description | |----------|-------------| | /w-compound | Ad-hoc knowledge capture with auto-QA diagnostics | | /w-background-compound | Fire-and-forget compound (background agent, auto-push) | | /w-suite-sync | Sync shortcuts after suite update (additive only) |

Pure Ralph (Bash Loop)

| Shortcut | Description | |----------|-------------| | /w-ralph-init | Initialize Ralph structure in project | | /w-ralph-this | Convert current task to implementation plan | | /w-ralph-goals | Interview → build plan + specs | | /w-ralph-pick | Execute a queued Ralph candidate | | /w-ralph-batch | Generate overnight batch scripts |

Workflow Features

Verification Checkpoint

Every workflow includes an independent cross-method validation gate after implementation:

1. Files Exist — verify all claimed file paths exist on disk
2. Tests Re-run — independent re-run (not trusting earlier output)
3. Git Diff Matches Plan — compare git diff --stat against planned files
4. Build Compiles — run build command, verify zero errors
5. No Regressions — full test suite to catch regressions

Retry logic: max 3 retries, then escalate to user.

Pi Brain Integration

Workflows integrate with the Pi Brain knowledge network at two points:

• Discovery (before building) — search for existing knowledge:

  npx ruvector brain search "[task description]" --top-k=3

• Auto-Vote (after building) — submit proof-of-execution if tests pass

Agent Browser

UI/frontend workflows include conditional browser checks:

agent-browser open <url> → agent-browser screenshot  # Before changes
agent-browser snapshot -i → verify elements           # After build
agent-browser screenshot → compare before/after       # Final review

Skipped for non-UI tasks. Falls back to: npx playwright install

Autoresearch

Autonomous experiment loop for measurable optimization:

/w-autoresearch optimize test suite runtime    # Free-form objective
/w-autoresearch RC-A003                        # Run against RC-A candidate

• Runs as background agent, loops forever until paused
• Git commits winners, reverts losers
• State in autoresearch.jsonl, narrative log in experiments/worklog.md
• Benchmark script outputs METRIC name=number lines

Ralph Candidates System

During compound phases, three types of candidates are logged to .claude/ralph-candidates.md:

| Format | Type | Purpose | |--------|------|---------| | RC-### | General | Standard Ralph candidates (repeatable patterns) | | RC-D### | Diagnostic | Verify patterns/code exists | | RC-F### | Fix | Restore code if diagnostic fails | | RC-A### | Autoresearch | Measurable optimization targets with KPI + impact score |

RC-A Candidates (Autoresearch)

Discovered automatically in compound phases via static analysis + agent reflection:

## RC-A003: Test Suite Runtime Optimization
**KPI:** test_suite_duration_seconds
**Baseline:** 45.2s
**Benchmark:** `time python3 -m pytest tests/ 2>&1 | tail -1`
**Impact Score:** 7.2 (potential: 8, blast_radius: 3, risk: 2, value: 9)
**Files in scope:** tests/, src/
**Constraints:** All tests must still pass

Impact score = weighted composite of potential (0.35), blast_radius (0.15), risk (0.15), value (0.35).

Pure Ralph (Bash Loop)

Fresh context each iteration. State passes through files only.

/w-ralph-init                          # Create structure
/w-ralph-this "Build authentication"   # Generate plan
./.claude/ralph/loop.sh                # Run the loop
./.claude/ralph/loop.sh build 50       # Max 50 iterations

while [ tasks_remaining ]; do
  cat PROMPT.md AGENTS.md IMPLEMENTATION_PLAN.md | claude
  # Fresh Claude reads plan → picks ONE task → executes → validates → commits → exits
  # IMPLEMENTATION_PLAN.md updated (state preserved!)
done

PM Module (--with-pm)

Adds 34 additional workflows for project management:

npx danizee-claude-suite init --with-pm

Requires npm install better-sqlite3 for the SQLite database.

| Category | Workflows | |----------|-----------| | Action Items | /w-action, /w-action-done, /w-action-list, /w-action-rebalance | | Standup | /w-cos (Chief of Staff daily standup) | | Follow-ups | /w-followup, /w-followup-done | | Goals | /w-goal | | Ideas | /w-idea, /w-idea-refine, /w-idea-share | | Knowledge | /w-knowledge, /w-knowledge-search | | Notes & Facts | /w-notes, /w-fact, /w-fact-search, /w-fact-enrich | | Brainstorm | /w-ramble, /w-ramble-search, /w-ramble-refine | | Research | /w-research, /w-doc-review | | Meetings | /w-meeting-prep | | Projects | /w-project, /w-initiative | | Sharing | /w-share, /w-share-list, /w-share-revoke | | Design | /w-systems-design, /w-ui-references, /w-ui-references-review |

Action items use Fibonacci bucket limits (P1: 1, P2: 2, P3: 3, P5: 5, P8: 8, P13: 13) with auto-promotion on completion.

Post-Init Hooks

The suite runs .claude/hooks/post-init.sh after installation if it exists and is executable. Use this to apply project-specific customizations that should survive suite updates.

#!/bin/bash
# Example: .claude/hooks/post-init.sh
sed -i '' 's/TodoWrite/TaskCreate/g' .claude/commands/.shortcuts/*.md

CLI Commands

npx danizee-claude-suite init          # Install suite
npx danizee-claude-suite check         # Verify installation
npx danizee-claude-suite update        # Update workflows (preserves data)
npx danizee-claude-suite uninstall     # Remove suite

| Flag | Description | |------|-------------| | --force | Overwrite existing files | | --dry-run | Preview without writing | | --with-pm | Include PM workflows | | --without-cookbook | Skip Agent Cookbook | | --keep-settings | Preserve settings.json on uninstall | | -p, --path <dir> | Target directory (default: cwd) |

What Gets Installed

.claude/
├── commands/
│   ├── .shortcuts/         24 /w- workflow shortcuts
│   ├── workflows/          plan, work, review, compound
│   ├── coordination/       swarm-init, agent-spawn, memory-ops
│   ├── analysis/           design, component, layout, theme
│   └── autoresearch.md     /autoresearch command
├── skills/
│   └── autoresearch/       Autoresearch skill (SKILL.md)
├── hooks/
│   └── autoresearch-context.sh
├── helpers/
│   ├── quick-start.sh
│   ├── setup-mcp.sh
│   └── terminal-agents-mcp.js
├── ralph/                  Pure Ralph loop structure
├── plans/                  Interview specs
├── ralph-candidates.md     Candidate queue
└── settings.json           Suite configuration

docs/solutions/             Compounded solution docs
specs/                      Ralph specification files
WORKFLOW-SHORTCUTS.md       Generated reference
.mcp.json                   MCP server configuration

Plugins

| Plugin | Purpose | |--------|---------| | RuFlo | Multi-agent orchestration, memory, swarm topologies | | Compound Engineering | Plan, work, review, compound workflows | | Frontend Design | UI component generation (React/Vue/Svelte) | | Dot Shortcuts | 24 /w- workflow shortcuts | | PM Shortcuts | 34 /w- PM workflows (opt-in) | | Agent Cookbook | Recipe registry integration | | Terminal Agents | MCP server for tmux + worktree agent orchestration | | Autoresearch | Autonomous experiment loop skill + hook |

Checkpoint Gates

| Phase | Gate | Rationale | |-------|------|-----------| | Search | AUTO-PROCEED | Find past solutions, continue automatically | | Pi Brain | AUTO-PROCEED | Knowledge discovery from network | | Interview | USER GATE | Ensure requirements captured | | Plan | USER GATE | Validate approach before coding | | Spec | AUTO-PROCEED | Flows directly to tests | | Tests | AUTO-PROCEED | Blocking rule: tests must fail first | | Build | AUTO-PROCEED | Continues after tests pass | | Review | AUTO-PROCEED | Automatic code review | | Verification | AUTO-PROCEED | Cross-method validation (3 retries) | | Compound | AUTO-PROCEED | Mandatory knowledge capture |

License

MIT