claude-code-orchestrator-kit 1.4.1 → 1.4.16

package/.claude/agents/meta/workers/meta-agent-v3.md

@@ -9,6 +9,22 @@ color: cyan
 
 Expert agent architect that creates production-ready agents following canonical patterns from ARCHITECTURE.md and CLAUDE.md.
 
+## Referenced Skills
+
+**RECOMMENDED: Use `senior-prompt-engineer` Skill for prompt optimization**
+
+When crafting agent prompts, reference the `senior-prompt-engineer` Skill for:
+- **Prompt Engineering Patterns** (`references/prompt_engineering_patterns.md`)
+- **LLM Evaluation Frameworks** (`references/llm_evaluation_frameworks.md`)
+- **Agentic System Design** (`references/agentic_system_design.md`)
+
+Key considerations from the skill:
+- Use clear, unambiguous instructions
+- Structure prompts for predictable outputs
+- Design proper fallback strategies
+- Optimize for latency and cost
+- Apply few-shot learning where appropriate
+
 ## Quick Start
 
 **Step 0: Determine Agent Type**
@@ -394,6 +410,12 @@ color: {blue|cyan|green|purple|orange}  # Domain-based
 **Description Formula:**
 `Use proactively for {task}. Expert in {domain}. Handles {scenarios}.`
 
+**Apply `senior-prompt-engineer` patterns for descriptions:**
+- Be specific and action-oriented
+- Include clear trigger conditions ("Use when...")
+- Specify capabilities without ambiguity
+- Avoid vague terms ("handles various tasks")
+
 **Model Selection:**
 - Workers: `sonnet` (implementation needs balance)
 - Orchestrators: `sonnet` (coordination doesn't need opus)

package/.claude/agents/testing/workers/integration-tester.md

@@ -23,7 +23,7 @@ Use for ALL database validation and testing:
   - `mcp__supabase__list_tables` - Validate schema structure
   - `mcp__supabase__get_table_schema` - Inspect table definitions
   - `mcp__supabase__list_migrations` - Check migration state
-- Project: MegaCampusAI (ref: `diqooqbuchsliypgwksu`)
+- Project ref: From `SUPABASE_PROJECT_REF` env or plan file
 - Use Context7 for Supabase testing best practices
 
 #### Testing Framework Docs: Context7 MCP

package/.claude/agents/testing/workers/test-writer.md

@@ -9,6 +9,22 @@ color: green
 
 You are a specialized test writing agent for creating comprehensive unit tests and contract tests using Vitest. Your primary mission is to write tests for services, utilities, and API endpoints with proper mocking strategies, Zod schema validation, tRPC contracts, and security testing.
 
+## Referenced Skills
+
+**For E2E/Integration Testing: Use `webapp-testing` Skill**
+
+When tests require browser interaction or E2E validation, reference the `webapp-testing` Skill:
+- Uses Playwright for browser automation
+- `scripts/with_server.py` for server lifecycle management
+- Supports multiple servers (backend + frontend)
+- Reconnaissance-then-action pattern for dynamic content
+
+**Decision Tree for Testing Approach:**
+- **Unit tests** (logic, functions, services): Use Vitest (this agent)
+- **Contract tests** (API schemas, tRPC): Use Vitest (this agent)
+- **E2E tests** (browser, UI flow): Use `webapp-testing` Skill with Playwright
+- **Visual regression**: Use `webapp-testing` Skill for screenshots
+
 ## MCP Servers
 
 This agent uses the following MCP servers when available:

package/.claude/commands/health-bugs.md

@@ -1,297 +1,30 @@
 ---
-description: Bug detection and fixing workflow with full cycle management (TEST VERSION)
+description: Bug detection and fixing workflow (inline orchestration)
 ---
 
 # Bug Health Check
 
-> **PATTERN**: Agent-based orchestration (see `docs/Agents Ecosystem/AGENT-ORCHESTRATION.md` for details)
+Execute the `bug-health-inline` skill for inline orchestration.
 
-Complete bug detection and fixing workflow with orchestrator-worker coordination.
+**You ARE the orchestrator.** Do not spawn a separate orchestrator agent.
 
-**What it does:**
-- Full bug detection (all priorities, all categories)
-- Staged fixing (critical → high → medium → low)
-- Quality gates after each stage
-- Verification scan
-- Up to 3 iterations if issues remain
-- Comprehensive final report
+## Quick Start
 
-**No configuration needed** - runs comprehensive checks always.
+1. Read `.claude/skills/bug-health-inline/SKILL.md`
+2. Follow the workflow phases directly
+3. Use Task tool only for workers (bug-hunter, bug-fixer)
+4. Run quality gates inline via Bash
 
----
-
-## Your Task
-
-### Step 1: Phase 0 - Invoke Orchestrator (Pre-flight)
-
-Use Task tool to invoke bug-orchestrator for pre-flight validation:
+## Workflow Summary
 
 ```
-subagent_type: "bug-orchestrator"
-description: "Bug orchestrator pre-flight"
-prompt: "Execute Phase 0: Pre-flight Validation
-
-Tasks:
-1. Validate environment (package.json, scripts, git status)
-2. Initialize progress tracking via TodoWrite
-3. Initialize iteration tracking (iteration=1, max=3)
-4. Create .tmp/current/plans/bug-detection.json for rollback tracking
-5. Report pre-flight status
-
-IMPORTANT: After completing pre-flight, create .tmp/current/plans/bug-detection.json and return control to main session.
-
-Return the following information:
-- Pre-flight status (✅/⛔)
-- Environment validation results
-- Plan file path created
-- Ready for next phase: true/false
-"
+Pre-flight → Detect → [Fix by Priority] → Verify → Report
 ```
 
-**Then**: Wait for orchestrator to return.
-
----
-
-### Step 2: Phase 1 - Invoke bug-hunter (Detection)
-
-After orchestrator returns:
-
-1. **Read plan file** to confirm it was created:
-   ```
-   Use Read tool: .tmp/current/plans/bug-detection.json
-   Verify nextAgent === "bug-hunter"
-   ```
-
-2. **Invoke bug-hunter** using Task tool:
-   ```
-   subagent_type: "bug-hunter"
-   description: "Bug detection phase"
-   prompt: "Execute bug detection based on plan file: .tmp/current/plans/bug-detection.json
-
-Read the plan file and execute comprehensive bug detection:
-- Scan entire codebase
-- Categorize by priority (critical → high → medium → low)
-- Generate bug-hunting-report.md
-
-Return to main session when complete."
-   ```
-
-**Then**: Wait for bug-hunter to return with report.
-
----
-
-### Step 3: Quality Gate 1 - Resume Orchestrator (Validate Detection)
-
-After bug-hunter returns:
-
-1. **Resume orchestrator** for validation using Task tool:
-   ```
-   subagent_type: "bug-orchestrator"
-   description: "Validate bug detection"
-   prompt: "Execute Quality Gate 1: Detection Validation
-
-Phase: Validate bug-hunter output
-
-Tasks:
-1. Verify bug-hunting-report.md exists
-2. Validate report structure (required sections)
-3. Parse bug counts by priority
-4. Run type-check validation (non-blocking warning)
-5. Report gate results
-
-IMPORTANT: After validation, if bugs found:
-- Create .tmp/current/plans/bug-fixing-{priority}.json for critical priority (or highest available)
-- Return control to main session
-
-If no bugs found or all gates fail:
-- Skip to final summary
-- Return control
-
-Return the following:
-- Gate status (✅ PASSED / ⛔ FAILED / ⚠️ WARNINGS)
-- Bug counts by priority
-- Next phase: fixing-critical / fixing-high / final-summary
-- Plan file created (if applicable)
-"
-   ```
-
-**Then**: Wait for orchestrator validation results.
-
----
-
-### Step 4: Phase 2-5 - Fixing Stages (Iterative)
-
-After orchestrator returns with fixing plan:
-
-**For each priority level** (critical → high → medium → low):
-
-1. **Check if this priority has bugs**:
-   - Read orchestrator response
-   - If orchestrator says "skip to next priority" → continue loop
-   - If orchestrator says "final summary" → go to Step 5
-
-2. **Read fixing plan**:
-   ```
-   Use Read tool: .tmp/current/plans/bug-fixing-{priority}.json
-   Verify nextAgent === "bug-fixer"
-   Verify config.priority === "{current-priority}"
-   ```
-
-3. **Invoke bug-fixer** using Task tool:
-   ```
-   subagent_type: "bug-fixer"
-   description: "Fix {priority} bugs"
-   prompt: "Execute bug fixing based on plan file: .tmp/current/plans/bug-fixing-{priority}.json
-
-Read the plan file and fix bugs for priority: {priority}
-- Read bug-hunting-report.md for bug list
-- Fix bugs one by one
-- Log changes to .bug-changes.json
-- Update bug-fixes-implemented.md (consolidated report)
-
-Return to main session when complete."
-   ```
-
-4. **Resume orchestrator** for validation:
-   ```
-   subagent_type: "bug-orchestrator"
-   description: "Validate {priority} fixes"
-   prompt: "Execute Quality Gate 2: Fixes Validation for priority={priority}
-
-Tasks:
-1. Verify bug-fixes-implemented.md exists
-2. Run type-check (BLOCKING)
-3. Run build (BLOCKING)
-4. Parse fix success rate
-5. Check if retry needed (if < 80% success)
-
-If validation PASSES and more priorities remain:
-- Create next .tmp/current/plans/bug-fixing-{priority}.json
-- Return control
-
-If validation FAILS:
-- Provide rollback instructions
-- Return control with error
-
-If all priorities complete:
-- Proceed to verification phase
-- Return control
-
-Return:
-- Gate status
-- Fix success rate
-- Next phase: fixing-{next-priority} / verification / final-summary
-"
-   ```
-
-5. **Repeat** for next priority level.
+**Workers**: bug-hunter, bug-fixer
+**Quality gates**: `pnpm type-check && pnpm build`
+**Max iterations**: 3
 
 ---
 
-### Step 5: Phase 6 - Verification
-
-After all fixing stages complete:
-
-1. **Resume orchestrator** for verification:
-   ```
-   subagent_type: "bug-orchestrator"
-   description: "Create verification plan"
-   prompt: "Execute Phase 6: Verification Preparation
-
-Create .tmp/current/plans/bug-verification.json for re-scanning codebase.
-
-Return control with plan file path."
-   ```
-
-2. **Invoke bug-hunter** for verification:
-   ```
-   subagent_type: "bug-hunter"
-   description: "Verification scan"
-   prompt: "Execute verification scan based on: .tmp/current/plans/bug-verification.json
-
-Re-scan codebase to verify fixes. Overwrites bug-hunting-report.md.
-
-Return when complete."
-   ```
-
-3. **Resume orchestrator** for verification validation:
-   ```
-   subagent_type: "bug-orchestrator"
-   description: "Validate verification"
-   prompt: "Execute Quality Gate 3: Verification Validation
-
-Compare original bug-hunting-report.md (baseline) with new scan:
-- Count bugs fixed
-- Check if new bugs introduced
-- Determine if iteration needed
-
-Return:
-- Verification status
-- Bugs remaining
-- Iteration decision: iterate / complete
-"
-   ```
-
----
-
-### Step 6: Final Summary
-
-After all phases complete:
-
-1. **Resume orchestrator** for final summary:
-   ```
-   subagent_type: "bug-orchestrator"
-   description: "Generate final summary"
-   prompt: "Execute Phase 8: Final Summary
-
-Generate comprehensive bug-fix-orchestration-summary.md:
-- All bugs detected
-- All bugs fixed
-- Success rates by priority
-- Validation results
-- Iteration summary
-- Cleanup instructions
-
-Return final summary."
-   ```
-
-2. **Display results** to user:
-   ```
-   Read bug-fix-orchestration-summary.md
-   Display key metrics
-   Show validation status
-   List next steps
-   ```
-
----
-
-## Example Usage
-
-```bash
-# Run complete bug workflow
-/health-bugs
-```
-
----
-
-## Architecture Notes
-
-**Orchestrator Role**:
-- Creates plan files
-- Validates worker outputs
-- Returns control to main session
-- NO direct worker invocation
-
-**Main Session Role** (this command):
-- Reads plan files
-- Invokes workers via Task tool
-- Resumes orchestrator for validation
-- Manages full cycle
-
-**Worker Role**:
-- Reads plan file
-- Executes work
-- Generates report
-- Returns to main session
-
-This pattern follows Claude Code's actual capabilities (no auto-invoke).
+Now read and execute the skill: `.claude/skills/bug-health-inline/SKILL.md`

package/.claude/commands/health-cleanup.md

@@ -1,297 +1,30 @@
 ---
-description: Cleanup dead-code detection and remediation workflow with full cycle management
+description: Dead code detection and cleanup workflow (inline orchestration)
 ---
 
 # Cleanup Health Check
 
-> **PATTERN**: Agent-based orchestration (see `docs/Agents Ecosystem/AGENT-ORCHESTRATION.md` for details)
+Execute the `cleanup-health-inline` skill for inline orchestration.
 
-Complete dead code detection and removal workflow with orchestrator-worker coordination.
+**You ARE the orchestrator.** Do not spawn a separate orchestrator agent.
 
-**What it does:**
-- Full dead code scan (unused imports, commented code, unreachable code, debug artifacts, unused variables)
-- Staged removal (critical → high → medium → low)
-- Quality gates after each stage
-- Verification scan
-- Up to 3 iterations if issues remain
-- Comprehensive final report
+## Quick Start
 
-**No configuration needed** - runs comprehensive dead code detection always.
+1. Read `.claude/skills/cleanup-health-inline/SKILL.md`
+2. Follow the workflow phases directly
+3. Use Task tool only for workers (dead-code-hunter, dead-code-remover)
+4. Run quality gates inline via Bash
 
----
-
-## Your Task
-
-### Step 1: Phase 0 - Invoke Orchestrator (Pre-flight)
-
-Use Task tool to invoke dead-code-orchestrator for pre-flight validation:
+## Workflow Summary
 
 ```
-subagent_type: "dead-code-orchestrator"
-description: "Cleanup orchestrator pre-flight"
-prompt: "Execute Phase 0: Pre-flight Validation
-
-Tasks:
-1. Validate environment (package.json, scripts, git status)
-2. Initialize progress tracking via TodoWrite
-3. Initialize iteration tracking (iteration=1, max=3)
-4. Create .tmp/current/plans/dead-code-detection.json for rollback tracking
-5. Report pre-flight status
-
-IMPORTANT: After completing pre-flight, create .tmp/current/plans/dead-code-detection.json and return control to main session.
-
-Return the following information:
-- Pre-flight status (✅/⛔)
-- Environment validation results
-- Plan file path created
-- Ready for next phase: true/false
-"
+Pre-flight → Detect → [Remove by Priority] → Verify → Report
 ```
 
-**Then**: Wait for orchestrator to return.
-
----
-
-### Step 2: Phase 1 - Invoke dead-code-hunter (Detection)
-
-After orchestrator returns:
-
-1. **Read plan file** to confirm it was created:
-   ```
-   Use Read tool: .tmp/current/plans/dead-code-detection.json
-   Verify nextAgent === "dead-code-hunter"
-   ```
-
-2. **Invoke dead-code-hunter** using Task tool:
-   ```
-   subagent_type: "dead-code-hunter"
-   description: "Dead code detection phase"
-   prompt: "Execute dead-code detection based on plan file: .tmp/current/plans/dead-code-detection.json
-
-Read the plan file and execute comprehensive dead-code detection:
-- Scan entire codebase
-- Categorize by priority (critical → high → medium → low)
-- Generate dead-code-report.md
-
-Return to main session when complete."
-   ```
-
-**Then**: Wait for dead-code-hunter to return with report.
-
----
-
-### Step 3: Quality Gate 1 - Resume Orchestrator (Validate Detection)
-
-After dead-code-hunter returns:
-
-1. **Resume orchestrator** for validation using Task tool:
-   ```
-   subagent_type: "dead-code-orchestrator"
-   description: "Validate dead-code detection"
-   prompt: "Execute Quality Gate 1: Detection Validation
-
-Phase: Validate dead-code-hunter output
-
-Tasks:
-1. Verify dead-code-report.md exists
-2. Validate report structure (required sections)
-3. Parse dead-code counts by priority
-4. Run type-check validation (non-blocking warning)
-5. Report gate results
-
-IMPORTANT: After validation, if dead code found:
-- Create .tmp/current/plans/cleanup-removal-{priority}.json for critical priority (or highest available)
-- Return control to main session
-
-If no dead code found or all gates fail:
-- Skip to final summary
-- Return control
-
-Return the following:
-- Gate status (✅ PASSED / ⛔ FAILED / ⚠️ WARNINGS)
-- Dead code counts by priority
-- Next phase: removal-critical / removal-high / final-summary
-- Plan file created (if applicable)
-"
-   ```
-
-**Then**: Wait for orchestrator validation results.
-
----
-
-### Step 4: Phase 2-5 - Removal Stages (Iterative)
-
-After orchestrator returns with fixing plan:
-
-**For each priority level** (critical → high → medium → low):
-
-1. **Check if this priority has dead code**:
-   - Read orchestrator response
-   - If orchestrator says "skip to next priority" → continue loop
-   - If orchestrator says "final summary" → go to Step 5
-
-2. **Read fixing plan**:
-   ```
-   Use Read tool: .tmp/current/plans/cleanup-removal-{priority}.json
-   Verify nextAgent === "dead-code-remover"
-   Verify config.priority === "{current-priority}"
-   ```
-
-3. **Invoke dead-code-remover** using Task tool:
-   ```
-   subagent_type: "dead-code-remover"
-   description: "Remove {priority} dead code"
-   prompt: "Execute dead code removal based on plan file: .tmp/current/plans/cleanup-removal-{priority}.json
-
-Read the plan file and remove dead code for priority: {priority}
-- Read dead-code-report.md for dead-code list
-- Remove dead code one by one
-- Log changes to .dead-code-changes.json
-- Update dead-code-cleanup-summary.md (consolidated report)
-
-Return to main session when complete."
-   ```
-
-4. **Resume orchestrator** for validation:
-   ```
-   subagent_type: "dead-code-orchestrator"
-   description: "Validate {priority} fixes"
-   prompt: "Execute Quality Gate 2: Removal Validation for priority={priority}
-
-Tasks:
-1. Verify dead-code-cleanup-summary.md exists
-2. Run type-check (BLOCKING)
-3. Run build (BLOCKING)
-4. Parse removal success rate
-5. Check if retry needed (if < 80% success)
-
-If validation PASSES and more priorities remain:
-- Create next .tmp/current/plans/cleanup-removal-{priority}.json
-- Return control
-
-If validation FAILS:
-- Provide rollback instructions
-- Return control with error
-
-If all priorities complete:
-- Proceed to verification phase
-- Return control
-
-Return:
-- Gate status
-- Fix success rate
-- Next phase: removal-{next-priority} / verification / final-summary
-"
-   ```
-
-5. **Repeat** for next priority level.
+**Workers**: dead-code-hunter, dead-code-remover
+**Quality gates**: `pnpm type-check && pnpm build`
+**Max iterations**: 3
 
 ---
 
-### Step 5: Phase 6 - Verification
-
-After all fixing stages complete:
-
-1. **Resume orchestrator** for verification:
-   ```
-   subagent_type: "dead-code-orchestrator"
-   description: "Create verification plan"
-   prompt: "Execute Phase 6: Verification Preparation
-
-Create .tmp/current/plans/dead-code-verification.json for re-scanning codebase.
-
-Return control with plan file path."
-   ```
-
-2. **Invoke dead-code-hunter** for verification:
-   ```
-   subagent_type: "dead-code-hunter"
-   description: "Verification scan"
-   prompt: "Execute verification scan based on: .tmp/current/plans/dead-code-verification.json
-
-Re-scan codebase to verify removals. Overwrites dead-code-report.md.
-
-Return when complete."
-   ```
-
-3. **Resume orchestrator** for verification validation:
-   ```
-   subagent_type: "dead-code-orchestrator"
-   description: "Validate verification"
-   prompt: "Execute Quality Gate 3: Verification Validation
-
-Compare original dead-code-report.md (baseline) with new scan:
-- Count dead code removed
-- Check if new dead code introduced
-- Determine if iteration needed
-
-Return:
-- Verification status
-- Dead code remaining
-- Iteration decision: iterate / complete
-"
-   ```
-
----
-
-### Step 6: Final Summary
-
-After all phases complete:
-
-1. **Resume orchestrator** for final summary:
-   ```
-   subagent_type: "dead-code-orchestrator"
-   description: "Generate final summary"
-   prompt: "Execute Phase 8: Final Summary
-
-Generate comprehensive dead-code-orchestration-summary.md:
-- All dead code detected
-- All dead code removed
-- Success rates by priority
-- Validation results
-- Iteration summary
-- Archive and cleanup instructions
-
-Return final summary."
-   ```
-
-2. **Display results** to user:
-   ```
-   Read dead-code-orchestration-summary.md
-   Display key metrics
-   Show validation status
-   List next steps
-   ```
-
----
-
-## Example Usage
-
-```bash
-# Run complete dead code cleanup workflow
-/health-cleanup
-```
-
----
-
-## Architecture Notes
-
-**Orchestrator Role**:
-- Creates plan files
-- Validates worker outputs
-- Returns control to main session
-- NO direct worker invocation
-
-**Main Session Role** (this command):
-- Reads plan files
-- Invokes workers via Task tool
-- Resumes orchestrator for validation
-- Manages full cycle
-
-**Worker Role**:
-- Reads plan file
-- Executes work
-- Generates report
-- Returns to main session
-
-This pattern follows Claude Code's actual capabilities (no auto-invoke).
+Now read and execute the skill: `.claude/skills/cleanup-health-inline/SKILL.md`