claudify-agent 1.0.1 → 1.0.3

package/.claude/skills/debug-optimizer/SKILL.md

@@ -1,399 +1,420 @@
-# SKILL.md: Debug Mode Optimizer
-
-## Description
-Automatic optimization of debug mode workflows for maximum efficiency. Activates automatically in debug mode to coordinate tools, MCPs, skills, hooks, and rules.
-
-## When to Use (Automatic Activation)
-**This skill activates AUTOMATICALLY in debug mode for ALL tasks.**
-
-**Purpose**: Coordinate all optimization mechanisms to minimize task completion time and token usage while maximizing quality.
-
----
-
-## Automatic Optimizations
-
-### 1. Tool Selection Optimization
-**Automatic Triggers**:
-```
-Task analysis → Select optimal tools based on:
-- Task type (exploration, implementation, validation)
-- Complexity (number of files, steps)
-- Required precision (exact vs conceptual)
-- Cost hierarchy (Tier 1 → Tier 4)
-```
-
-**Optimization Rules**:
-```
-✓ ALWAYS start with Tier 1 tools (Read, Glob, Grep)
-✓ Escalate one tier at a time only if needed
-✓ NEVER use expensive tools when cheap tools work
-✓ Track tool effectiveness for future optimization
-```
-
----
-
-### 2. MCP Server Optimization
-**Automatic Triggers**:
-```
-File patterns + Keywords + Task type → Select optimal MCPs:
-- components/**, "UI" → cursor-ide-browser
-- __tests__/**, "E2E" → user-playwright
-- "PR", "issue" → user-github
-- "lint" → user-ESLint
-- "docs", "how to" → user-context7
-- "performance" → user-chrome-devtools
-```
-
-**Optimization Rules**:
-```
-✓ ALWAYS run user-ESLint after code changes (cheap, high value)
-✓ ALWAYS fetch user-context7 before library implementation (prevents errors)
-✓ Use cursor-ide-browser ONLY when visual verification essential
-✓ Chain MCPs automatically for complex workflows
-✓ Track MCP usage and cost for optimization
-```
-
----
-
-### 3. Parallel Subagent Optimization
-**Automatic Triggers**:
-```
-Task complexity analysis → Launch parallel subagents:
-- >3 areas mentioned → Launch 4 subagents
-- >10 steps → Launch 3 subagents
-- Bug investigation → Launch 4 subagents
-- Large refactor (>5 files) → Launch 4 subagents
-```
-
-**Configuration** (Automatic):
-```yaml
-model: "fast"  # Cost optimization
-readonly: true  # Safety
-run_in_background: true  # Non-blocking
-timeout: 120s  # Time-box per subagent
-max_parallel: 4  # Diminishing returns after 4
-```
-
----
-
-### 4. Context Engineering Optimization
-**Automatic Triggers**:
-```
-Task keywords + File patterns → Load relevant Tier 2 context:
-- "test" → Testing context
-- "API" → API development context
-- "UI", "component" → UI/Styling context
-- "performance" → Performance context
-- "database" → Database context
-- "security" → Security context
-```
-
-**Optimization Rules**:
-```
-✓ Load ONLY relevant context (avoid over-specification)
-✓ Use Tier 1 (minimal) + Tier 2 (task-specific) injection
-✓ Prune context after task completion
-✓ Agent discovers details via search tools (not context)
-```
-
----
-
-### 5. Hook-Based Validation Optimization
-**Automatic Triggers**:
-```
-After code changes (after_code_change hook):
-- auto-lint-fix → Fix lint issues first
-- auto-validate → Unified validation (lint + typecheck + tests + MCP)
-
-After plan mode (plan_mode_exit hook):
-- plan-quality-tracker → Track metrics and provide feedback
-```
-
-**Validation Pipeline** (Automatic):
-```
-1. Code change detected
-2. after_code_change hook triggers
-3. auto-lint-fix: Run ESLint, auto-fix
-4. auto-validate: 
-   - Lint check
-   - Typecheck
-   - Run tests
-   - Validate MCP tools
-5. Report results to user
-```
-
----
-
-## Debug-Specific Workflows
-
-### Workflow 1: Feature Implementation (Optimized)
-```
-User: "Add user authentication with JWT"
-
-Automatic Optimizations:
-1. **Parallel Exploration** (4 subagents, 60s):
-   - Find auth patterns
-   - Find JWT libraries
-   - Find existing auth code
-   - Find auth tests
-
-2. **Context Loading** (Tier 2):
-   - Security context
-   - API context
-
-3. **MCP Chain**:
-   - user-context7: Fetch JWT library docs
-   - Implement feature
-   - user-ESLint: Auto-lint
-   - user-github: Create PR
-
-4. **Validation Hooks**:
-   - auto-lint-fix: Fix issues
-   - auto-validate: Unified validation (includes MCP validation + tests)
-
-Time: 5-7 minutes (vs 15-20 manual)
-```
-
----
-
-### Workflow 2: Bug Fix (Optimized)
-```
-User: "Fix checkout bug #456"
-
-Automatic Optimizations:
-1. **Parallel Exploration** (4 subagents, 90s):
-   - Find checkout code
-   - Find payment code
-   - Find recent changes
-   - Find similar issues
-
-2. **Context Loading** (Tier 2):
-   - API context (if API-related)
-   - Database context (if DB-related)
-
-3. **MCP Chain**:
-   - user-github: Fetch issue #456
-   - user-context7: Check payment docs (if needed)
-   - Implement fix
-   - user-ESLint: Auto-lint
-   - user-github: Create PR
-
-4. **Validation Hooks**:
-   - auto-lint-fix: Fix issues
-   - auto-validate: Unified validation (includes MCP validation + tests)
-
-Time: 3-5 minutes (vs 10-15 manual)
-```
-
----
-
-### Workflow 3: Large Refactor (Optimized)
-```
-User: "Migrate from Prisma to Drizzle"
-
-Automatic Optimizations:
-1. **Parallel Exploration** (4 subagents, 120s):
-   - Find all Prisma usage
-   - Find all queries
-   - Find all migrations
-   - Find all tests
-
-2. **Context Loading** (Tier 2):
-   - Database context
-   - Testing context
-
-3. **MCP Chain**:
-   - user-context7: Fetch Drizzle docs
-   - Implement migration (incremental)
-   - user-ESLint: Auto-lint after each change
-   - test-runner: Run tests after each change
-
-4. **Validation Hooks** (after each step):
-   - auto-lint-fix: Fix issues
-   - auto-validate: Run unified validation (includes tests)
-
-Time: 20-30 minutes (vs 60-90 manual)
-```
-
----
-
-## Error Pattern Recognition
-
-### Automatic Error Detection
-```
-Monitor for error patterns:
-- ESLint errors → Trigger auto-lint-fix
-- Test failures → Trigger auto-validate with debug mode
-- Build errors → Fetch build docs via user-context7
-- Runtime errors → Trigger browser debugging
-```
-
-### Automatic Recovery
-```
-Error detected → Automatic response:
-1. Identify error type
-2. Fetch relevant docs (user-context7)
-3. Implement fix
-4. Validate (lint + tests)
-5. If still failing → Escalate to human
-```
-
----
-
-## Performance Metrics (Automatic Tracking)
-
-### Track Per Task
-```
-- Task completion time
-- Token consumption
-- Tool calls count
-- MCP calls count
-- Validation gate pass rate
-- User satisfaction (manual rating)
-```
-
-### Weekly Optimization Review
-```
-Analyze metrics:
-- Which optimizations saved most time?
-- Which tools/MCPs most effective?
-- Where are bottlenecks?
-- What can be further automated?
-- Adjust rules and triggers accordingly
-```
-
----
-
-## Cost Optimization (Enforced)
-
-### Token Budget Allocation
-```
-Tier 1 tools (Read, Glob, Grep): 40% of budget
-Tier 2 tools (SemanticSearch, StrReplace): 30% of budget
-Tier 3 tools (Task, Shell, ReadLints): 20% of budget
-Tier 4 tools (Parallel subagents, Browser MCP): 10% of budget
-```
-
-### Automatic Cost Control
-```
-If task exceeds budget:
-1. Warn user
-2. Suggest scope reduction
-3. Offer to continue with approval
-```
-
----
-
-## Examples
-
-### Example 1: Add Dark Mode (Optimized)
-User: "Add dark mode toggle to settings page"
-
-**Automatic Optimizations**:
-```
-1. Parallel subagents (2, 45s):
-   - "Find settings page structure"
-   - "Find Tailwind dark mode examples"
-
-2. Context loaded: UI/Styling
-
-3. MCP chain:
-   - user-context7: Tailwind dark mode docs
-   - cursor-ide-browser: Verify UI
-   - user-ESLint: Auto-lint
-
-4. Hooks:
-   - auto-lint-fix: ✅ Passed
-   - auto-validate: ✅ All validations passed
-
-Result: 4 minutes, 100% quality gates pass
-```
-
----
-
-### Example 2: Performance Audit (Optimized)
-User: "Audit and improve dashboard performance"
-
-**Automatic Optimizations**:
-```
-1. Parallel subagents (3, 90s):
-   - "Find dashboard code"
-   - "Find performance bottlenecks"
-   - "Find optimization patterns"
-
-2. Context loaded: Performance
-
-3. MCP chain:
-   - user-chrome-devtools: Lighthouse audit
-   - cursor-ide-browser: CPU profiling
-   - user-context7: React optimization docs
-   - Implement optimizations
-   - user-chrome-devtools: Re-audit
-
-4. Hooks:
-   - auto-validate: ✅ All validations passed (MCP + tests)
-
-Result: 8 minutes, Lighthouse score 92 (from 67)
-```
-
----
-
-### Example 3: Security Review (Optimized)
-User: "Review authentication for vulnerabilities"
-
-**Automatic Optimizations**:
-```
-1. Parallel subagents (4, 90s):
-   - "Find auth code"
-   - "Find token validation"
-   - "Find password handling"
-   - "Find input validation"
-
-2. Context loaded: Security
-
-3. MCP chain:
-   - user-context7: OWASP auth guidelines
-   - Implement security fixes
-   - user-ESLint: Auto-lint
-   - test-runner: Run security tests
-
-4. Hooks:
-   - auto-lint-fix: ✅ Fixed 3 issues
-   - test-runner: ✅ All tests pass
-
-Result: 6 minutes, 5 vulnerabilities fixed
-```
-
----
-
-## Anti-Patterns (Automatically Avoided)
-
-### ❌ Over-Automation
-**Prevention**: Start with minimal automation, add only when pain points observed
-
-### ❌ Context Bloat
-**Prevention**: Enforce 20-line max per context file, use references over content
-
-### ❌ MCP Cost Explosion
-**Prevention**: Cost-awareness rules, prefer cheaper tools first, track usage
-
-### ❌ Hook Failures Blocking Progress
-**Prevention**: Hooks exit gracefully, agent continues if hooks fail
-
----
-
-## Integration with Other Skills
-
-### Works with:
-- **mcp-mastery**: Coordinates MCP selection
-- **tool-selection**: Optimizes tool choices
-- **parallel-exploration**: Launches subagents efficiently
-
-### Coordinates with:
-- **Rules**: Enforces tool-auto-selection, parallel-exploration-auto, context tiers
-- **Hooks**: Triggers validation pipeline
-- **MCPs**: Chains multiple servers for complex workflows
-
----
-
-**Note**: This skill activates automatically in debug mode. No manual intervention needed. All optimizations applied transparently to maximize efficiency.
+---
+name: debug-optimizer
+description: Coordinate tools, MCP servers, subagents and validation hooks for debugging and multi-step tasks. Use when investigating a bug, a failing test, or a task spanning many files.
+---
+
+# SKILL.md: Debug Optimizer
+
+## Description
+Automatic optimization of debugging and multi-step workflows for maximum efficiency. Coordinates tools, MCPs, skills, hooks, and rules.
+
+## When to Use (Automatic Activation)
+**This skill activates AUTOMATICALLY for debugging and multi-step tasks.**
+
+**Purpose**: Coordinate all optimization mechanisms to minimize task completion time and token usage while maximizing quality.
+
+---
+
+## Root Cause Debugging Protocol
+
+1. **Reproduce & Isolate:** Recreate the exact failure with a minimal reproduction or a single targeted test.
+2. **Inspect & Trace:** Read actual runtime logs, stack traces, and variable state instead of guessing.
+3. **Hypothesize & Form Minimal Fix:** Identify the root mechanism of the failure. Make the smallest diff that fully fixes it.
+4. **Regression Prevention:** Add a targeted regression test proving the bug is gone and adjacent modules still work.
+
+**Never**:
+- ❌ Wrap code in `try-catch` without handling or logging the error
+- ❌ Silence TypeScript errors with `@ts-ignore` or `as any`
+- ❌ Write sprawling test fixtures when a focused unit test suffices
+
+---
+
+## Automatic Optimizations
+
+### 1. Tool Selection Optimization
+**Automatic Triggers**:
+```
+Task analysis → Select optimal tools based on:
+- Task type (exploration, implementation, validation)
+- Complexity (number of files, steps)
+- Required precision (exact vs conceptual)
+- Cost hierarchy (Tier 1 → Tier 4)
+```
+
+**Optimization Rules**:
+```
+✓ ALWAYS start with Tier 1 tools (Read, Glob, Grep)
+✓ Escalate one tier at a time only if needed
+✓ NEVER use expensive tools when cheap tools work
+✓ Track tool effectiveness for future optimization
+```
+
+---
+
+### 2. MCP Server Optimization
+**Automatic Triggers**:
+```
+File patterns + Keywords + Task type → Select optimal MCPs:
+- components/**, "UI" → claude-in-chrome
+- __tests__/**, "E2E" → playwright
+- "PR", "issue" → github
+- "lint" → eslint
+- "docs", "how to" → context7
+- "performance" → chrome-devtools
+```
+
+**Optimization Rules**:
+```
+✓ ALWAYS run eslint after code changes (cheap, high value)
+✓ ALWAYS fetch context7 before library implementation (prevents errors)
+✓ Use claude-in-chrome ONLY when visual verification essential
+✓ Chain MCPs automatically for complex workflows
+✓ Track MCP usage and cost for optimization
+```
+
+---
+
+### 3. Parallel Subagent Optimization
+**Automatic Triggers**:
+```
+Task complexity analysis → Launch parallel subagents:
+- >3 areas mentioned → Launch 4 subagents
+- >10 steps → Launch 3 subagents
+- Bug investigation → Launch 4 subagents
+- Large refactor (>5 files) → Launch 4 subagents
+```
+
+**Configuration** (Automatic):
+```yaml
+model: "haiku"  # Cost optimization
+run_in_background: true  # Non-blocking
+timeout: 120s  # Time-box per subagent
+max_parallel: 4  # Diminishing returns after 4
+```
+
+---
+
+### 4. Context Engineering Optimization
+**Automatic Triggers**:
+```
+Task keywords + File patterns → Load relevant Tier 2 context:
+- "test" → Testing context
+- "API" → API development context
+- "UI", "component" → UI/Styling context
+- "performance" → Performance context
+- "database" → Database context
+- "security" → Security context
+```
+
+**Optimization Rules**:
+```
+✓ Load ONLY relevant context (avoid over-specification)
+✓ Use Tier 1 (minimal) + Tier 2 (task-specific) injection
+✓ Prune context after task completion
+✓ Agent discovers details via search tools (not context)
+```
+
+---
+
+### 5. Hook-Based Validation Optimization
+**Automatic Triggers**:
+```
+After each Edit/Write (PostToolUse hook):
+- auto-lint-fix → Fix lint issues in the edited file
+
+When Claude finishes a response (Stop hook):
+- auto-validate → Unified validation (lint + typecheck + tests + MCP)
+- auto-security-review → Scan written code for vulnerabilities
+- plan-quality-tracker → Track metrics and provide feedback
+
+At session start (SessionStart hook):
+- auto-learning-review → Weekly learning report from plan metrics
+```
+
+**Validation Pipeline** (Automatic):
+```
+1. Edit/Write → PostToolUse hook → auto-lint-fix (ESLint --fix on that file)
+2. Claude tries to stop → Stop hook → auto-validate:
+   - Lint check
+   - Typecheck
+   - Run tests
+   - Validate MCP tools
+3. Failures block the stop and are sent back to Claude to fix
+4. Report results to user
+```
+
+---
+
+## Debug-Specific Workflows
+
+### Workflow 1: Feature Implementation (Optimized)
+```
+User: "Add user authentication with JWT"
+
+Automatic Optimizations:
+1. **Parallel Exploration** (4 subagents, 60s):
+   - Find auth patterns
+   - Find JWT libraries
+   - Find existing auth code
+   - Find auth tests
+
+2. **Context Loading** (Tier 2):
+   - Security context
+   - API context
+
+3. **MCP Chain**:
+   - context7: Fetch JWT library docs
+   - Implement feature
+   - eslint: Auto-lint
+   - github: Create PR
+
+4. **Validation Hooks**:
+   - auto-lint-fix: Fix issues
+   - auto-validate: Unified validation (includes MCP validation + tests)
+
+Time: 5-7 minutes (vs 15-20 manual)
+```
+
+---
+
+### Workflow 2: Bug Fix (Optimized)
+```
+User: "Fix checkout bug #456"
+
+Automatic Optimizations:
+1. **Parallel Exploration** (4 subagents, 90s):
+   - Find checkout code
+   - Find payment code
+   - Find recent changes
+   - Find similar issues
+
+2. **Context Loading** (Tier 2):
+   - API context (if API-related)
+   - Database context (if DB-related)
+
+3. **MCP Chain**:
+   - github: Fetch issue #456
+   - context7: Check payment docs (if needed)
+   - Implement fix
+   - eslint: Auto-lint
+   - github: Create PR
+
+4. **Validation Hooks**:
+   - auto-lint-fix: Fix issues
+   - auto-validate: Unified validation (includes MCP validation + tests)
+
+Time: 3-5 minutes (vs 10-15 manual)
+```
+
+---
+
+### Workflow 3: Large Refactor (Optimized)
+```
+User: "Migrate from Prisma to Drizzle"
+
+Automatic Optimizations:
+1. **Parallel Exploration** (4 subagents, 120s):
+   - Find all Prisma usage
+   - Find all queries
+   - Find all migrations
+   - Find all tests
+
+2. **Context Loading** (Tier 2):
+   - Database context
+   - Testing context
+
+3. **MCP Chain**:
+   - context7: Fetch Drizzle docs
+   - Implement migration (incremental)
+   - eslint: Auto-lint after each change
+   - auto-validate: Run tests after each change
+
+4. **Validation Hooks** (after each step):
+   - auto-lint-fix: Fix issues
+   - auto-validate: Run unified validation (includes tests)
+
+Time: 20-30 minutes (vs 60-90 manual)
+```
+
+---
+
+## Error Pattern Recognition
+
+### Automatic Error Detection
+```
+Monitor for error patterns:
+- ESLint errors → Trigger auto-lint-fix
+- Test failures → auto-validate blocks the stop and feeds failures back
+- Build errors → Fetch build docs via context7
+- Runtime errors → Trigger browser debugging
+```
+
+### Automatic Recovery
+```
+Error detected → Automatic response:
+1. Identify error type
+2. Fetch relevant docs (context7)
+3. Implement fix
+4. Validate (lint + tests)
+5. If still failing → Escalate to human
+```
+
+---
+
+## Performance Metrics (Automatic Tracking)
+
+### Track Per Task
+```
+- Task completion time
+- Token consumption
+- Tool calls count
+- MCP calls count
+- Validation gate pass rate
+- User satisfaction (manual rating)
+```
+
+### Weekly Optimization Review
+```
+Analyze metrics:
+- Which optimizations saved most time?
+- Which tools/MCPs most effective?
+- Where are bottlenecks?
+- What can be further automated?
+- Adjust rules and triggers accordingly
+```
+
+---
+
+## Cost Optimization (Enforced)
+
+### Token Budget Allocation
+```
+Tier 1 tools (Read, Glob, Grep): 40% of budget
+Tier 2 tools (Explore (quick), Edit): 30% of budget
+Tier 3 tools (Agent, Bash, Lint check): 20% of budget
+Tier 4 tools (Parallel subagents, Browser MCP): 10% of budget
+```
+
+### Automatic Cost Control
+```
+If task exceeds budget:
+1. Warn user
+2. Suggest scope reduction
+3. Offer to continue with approval
+```
+
+---
+
+## Examples
+
+### Example 1: Add Dark Mode (Optimized)
+User: "Add dark mode toggle to settings page"
+
+**Automatic Optimizations**:
+```
+1. Parallel subagents (2, 45s):
+   - "Find settings page structure"
+   - "Find Tailwind dark mode examples"
+
+2. Context loaded: UI/Styling
+
+3. MCP chain:
+   - context7: Tailwind dark mode docs
+   - claude-in-chrome: Verify UI
+   - eslint: Auto-lint
+
+4. Hooks:
+   - auto-lint-fix: ✅ Passed
+   - auto-validate: ✅ All validations passed
+
+Result: 4 minutes, 100% quality gates pass
+```
+
+---
+
+### Example 2: Performance Audit (Optimized)
+User: "Audit and improve dashboard performance"
+
+**Automatic Optimizations**:
+```
+1. Parallel subagents (3, 90s):
+   - "Find dashboard code"
+   - "Find performance bottlenecks"
+   - "Find optimization patterns"
+
+2. Context loaded: Performance
+
+3. MCP chain:
+   - chrome-devtools: Lighthouse audit
+   - chrome-devtools: CPU profiling
+   - context7: React optimization docs
+   - Implement optimizations
+   - chrome-devtools: Re-audit
+
+4. Hooks:
+   - auto-validate: ✅ All validations passed (MCP + tests)
+
+Result: 8 minutes, Lighthouse score 92 (from 67)
+```
+
+---
+
+### Example 3: Security Review (Optimized)
+User: "Review authentication for vulnerabilities"
+
+**Automatic Optimizations**:
+```
+1. Parallel subagents (4, 90s):
+   - "Find auth code"
+   - "Find token validation"
+   - "Find password handling"
+   - "Find input validation"
+
+2. Context loaded: Security
+
+3. MCP chain:
+   - context7: OWASP auth guidelines
+   - Implement security fixes
+   - eslint: Auto-lint
+   - auto-security-review: Scan written code
+
+4. Hooks:
+   - auto-lint-fix: ✅ Fixed 3 issues
+   - auto-validate: ✅ All tests pass
+
+Result: 6 minutes, 5 vulnerabilities fixed
+```
+
+---
+
+## Anti-Patterns (Automatically Avoided)
+
+### ❌ Over-Automation
+**Prevention**: Start with minimal automation, add only when pain points observed
+
+### ❌ Context Bloat
+**Prevention**: Enforce 20-line max per context file, use references over content
+
+### ❌ MCP Cost Explosion
+**Prevention**: Cost-awareness rules, prefer cheaper tools first, track usage
+
+### ❌ Hook Failures Blocking Progress
+**Prevention**: Hooks exit gracefully, agent continues if hooks fail
+
+---
+
+## Integration with Other Skills
+
+### Works with:
+- **mcp-mastery**: Coordinates MCP selection
+- **tool-selection**: Optimizes tool choices
+- **parallel-exploration**: Launches subagents efficiently
+
+### Coordinates with:
+- **Rules**: Enforces tool-auto-selection, parallel-exploration-auto, context tiers
+- **Hooks**: Triggers validation pipeline
+- **MCPs**: Chains multiple servers for complex workflows
+
+---
+
+**Note**: This skill activates automatically for debugging and multi-step tasks. No manual intervention needed. All optimizations applied transparently to maximize efficiency.