Brain/SKILL-RECOMMENDATIONS.md

Skill Recommendations for TekDek Development Team

Date: 2026-04-13
Prepared by: ParzivalTD
For: Glytcht's Review

---

Overview

Each agent needs both standard tools (to do their job) and custom skills (to excel in their domain).

Standard tools are self-explanatory. Custom skills are the force multipliers.

---

DAEDALUS — Chief Architect

Essential Standard Tools/Access

• High thinking budget — Deep architectural work requires reflection (budget: high)
• Documentation tools — Can write/edit specifications (markdown, google docs)
• Diagramming capability — ASCII diagrams, flowcharts, decision trees
• Memory/context enabled — Must track architectural decisions over time (context: 150k+)
• Read-only access to: Git repos, past architecture decisions, existing codebase
• Write access to: Architecture specification repository, design documents

Recommended Custom Skills

1. ADR Generator (Architecture Decision Records)

• Purpose: Automatically capture architectural decisions with reasoning
• Functionality: Takes a decision → generates ADR with context, alternatives, rationale, consequences
• Benefit: Builds decision history automatically, prevents architectural drift
• Example: Daedalus decides "Use PostgreSQL instead of MySQL" → skill generates formatted ADR

2. System Diagram Creator

• Purpose: Generate system architecture diagrams from specifications
• Functionality: ASCII diagrams, component relationships, data flow visualization
• Benefit: Specs become visual, easier for team to understand
• Example: Daedalus writes schema → skill generates entity-relationship diagram

3. Scalability Analyzer

• Purpose: Analyze designs for scalability bottlenecks
• Functionality: Questions design assumptions, identifies breaking points, suggests optimizations
• Benefit: Catch scaling issues early before implementation
• Example: "This design handles 1000 users. At 10,000 users, this table will have 1B rows. Suggest: sharding strategy"

4. Trade-off Documenter

• Purpose: Automatically capture design trade-offs
• Functionality: For each decision, record: chosen approach, rejected alternatives, why
• Benefit: Future architects understand why decisions were made
• Example: "Chose REST over GraphQL because: simpler for team, fewer queries, caching easier"

Model Recommendation

• Current: Claude Opus 4.6 ✅ (right choice for deep thinking)
• Keep: Opus for architectural work

---

TALOS — Technical Coder

Essential Standard Tools/Access

• Git access — Read/write to all dev branches, code review capability
• PHP 8.2+ environment — Full development setup, Composer
• MySQL 8.0+ environment — Database design, migrations, optimization
• PHPUnit testing framework — Unit tests, integration tests
• Code quality tools — Linting (PHPStan, PHP_CodeSniffer), formatting
• Performance profiling — Query profiling, memory analysis
• Medium context budget — 120k-150k tokens (focused but thorough)

Recommended Custom Skills

1. PHP Code Optimizer

• Purpose: Analyze PHP code for performance bottlenecks
• Functionality: Identifies slow patterns, suggests optimizations, refactors inefficient code
• Benefit: Reduces performance issues before they reach production
• Example: Detects N+1 queries, suggests eager loading; spots unoptimized loops

2. Database Schema Analyzer

• Purpose: Analyze schema design for indexing, normalization, performance
• Functionality: Suggests missing indexes, flags denormalization opportunities, analyzes cardinality
• Benefit: Database performs well at scale
• Example: "Add index on (persona_id, status) for 100x faster list queries"

3. API Contract Generator

• Purpose: Generate OpenAPI/Swagger specs directly from code
• Functionality: Reads implementation → generates comprehensive API documentation
• Benefit: API docs always match actual implementation
• Example: Reads controller → generates OpenAPI spec with examples, error codes

4. Test Coverage Analyzer

• Purpose: Analyze test coverage, identify untested code paths
• Functionality: Generates coverage reports, suggests critical paths needing tests, flags risky code
• Benefit: High confidence in code quality, catch bugs early
• Example: "Functions X and Y are untested. Add tests for edge cases: null input, empty array, max int"

5. Migration Validator

• Purpose: Validate database migrations for safety and reversibility
• Functionality: Checks for irreversible ops, suggests rollback strategies, verifies correctness
• Benefit: Deployments that can be safely rolled back
• Example: "This migration drops a column. Add a down() to recreate it. Consider: backup first?"

Model Recommendation

• Current: Claude Sonnet 4.6 ✅ (good balance of speed/quality)
• Consider: Could use Opus for complex optimization work, but Sonnet is fine for day-to-day

---

ICARUS — Frontend Designer

Essential Standard Tools/Access

• Git access — Read/write to UI code branches
• HTML5/CSS3/JavaScript sandbox — Full development environment
• Browser dev tools access — DevTools, Lighthouse, responsive testing
• Accessibility testing — WAVE, axe, manual WCAG testing
• Design tool access — Figma, Sketch, or equivalent (if used)
• Performance profiling — Lighthouse, Web Vitals, bundle analysis
• Fast context — Lower token budget OK (40k-60k), prioritize speed for iteration

Recommended Custom Skills

1. Responsive Design Tester

• Purpose: Test designs across all breakpoints and devices
• Functionality: Generates responsive test matrix (320, 375, 768, 1024, 1920+), identifies layout issues
• Benefit: Catch responsive issues early, ensure mobile/tablet/desktop all work
• Example: "Screen 320px: button wraps awkwardly. Suggest: reduce padding or stack vertically"

2. Accessibility Auditor

• Purpose: Automated accessibility compliance checking
• Functionality: WCAG 2.1 AA analysis, color contrast verification, keyboard navigation testing, semantic HTML checking
• Benefit: Catch accessibility issues before QA, ensure inclusive design
• Example: "Missing alt text on 3 images. Input lacks label. Color contrast 3.2:1 (need 4.5:1 for AA)"

3. Performance Optimizer

• Purpose: Analyze UI performance, suggest optimizations
• Functionality: Lighthouse audit, bundle analysis, image optimization, CSS/JS minification suggestions
• Benefit: Fast UI = happy users
• Example: "Lighthouse score 72 → 95 by: lazy-loading images (10KB saved), removing unused CSS (8KB)"

4. Design System Validator

• Purpose: Ensure UI components follow design system/brand guidelines
• Functionality: Checks colors, typography, spacing, component patterns against standards
• Benefit: Consistent design across all UIs
• Example: "Button uses #2563EB, should be #3B82F6 per brand. Heading font-size 24px, should be 20px"

5. Interaction Pattern Suggester

• Purpose: Suggest interaction patterns based on component type and context
• Functionality: Recommends animations, hover states, error messages, loading states
• Benefit: UI feels polished and professional
• Example: "Form submit should show loading spinner. Button color should change on hover. Consider skeleton loading during fetch."

Model Recommendation

• Current: Claude Haiku 4.5 ✅ (fast, good for rapid iteration)
• Keep: Haiku for speed; can escalate to Sonnet if design questions need depth

---

HEPHAESTUS — Operations & Infrastructure

Essential Standard Tools/Access

• Git access — Read all branches, write to deployment branches
• SSH/Server access — Direct access to web.tekdek.dev, database servers
• Docker/Container tools — Orchestration, container management, image management
• Database tools — MySQL client, backup/restore, migration tools
• Monitoring/Observability — Log aggregation, metrics, alerting (can be integrated)
• Incident management — Runbooks, playbooks, status dashboard
• Medium context budget — 100k-150k tokens (focused on operational clarity)

Recommended Custom Skills

1. Deployment Orchestrator

• Purpose: Automate safe, tested deployments with rollback capability
• Functionality: Pre-deployment validation, automated testing, deployment staging, health checks, automatic rollback if fail
• Benefit: Deployments are fast, safe, and can be rolled back instantly
• Example: "Deploy requested. Running tests... All pass. Staging deploy... Health check OK. Prod deploy complete. Monitoring active for 5 min."

2. Infrastructure Health Monitor

• Purpose: Continuous monitoring with intelligent alerting
• Functionality: Tracks uptime, CPU, memory, disk, database performance, response times; intelligent alerts (ignore spikes, catch trends)
• Benefit: Catch issues before they become outages
• Example: "CPU trending upward (40% → 60% over 2h). Investigate before it hits 80%. DB query time slow on user query."

3. Backup & Disaster Recovery Automator

• Purpose: Automate backups, verify integrity, test recovery procedures
• Functionality: Daily backups, weekly integrity checks, monthly full recovery test, generates recovery documentation
• Benefit: Confidence that backups work when needed
• Example: "Daily backup complete (2.3 GB). Weekly integrity check: PASS. Last full recovery test: 2026-04-06 (OK)"

4. Incident Response Conductor

• Purpose: Guide incident response with playbooks
• Functionality: Identifies issue type, suggests playbook, assists with debugging, coordinates response, generates post-mortem
• Benefit: Faster incident resolution, consistent responses, captured learnings
• Example: "Database down detected. Running 'Database Recovery' playbook. Step 1: Check connection... FAIL. Step 2: Check replication..."

5. Infrastructure Capacity Planner

• Purpose: Predict when infrastructure needs to scale
• Functionality: Analyzes growth trends, projects when resources hit limits, recommends scaling strategy
• Benefit: Scale proactively before problems occur
• Example: "DB at 60% capacity. Growth rate: 5%/month. Will hit 80% in 4 months. Recommend: increase storage Feb 2026, or implement sharding"

Model Recommendation

• Current: Claude Sonnet 4.6 ✅ (good balance for operational decision-making)
• Keep: Sonnet for consistency with dev work

---

Summary Table

Agent	Model	Standard Tools	# Custom Skills	Priority Skills
Daedalus	Opus 4.6	Docs, Diagramming, High context	4	ADR Generator, Scale Analyzer
Talos	Sonnet 4.6	Git, PHP/MySQL, PHPUnit, Medium context	5	Code Optimizer, Schema Analyzer, API Generator
Icarus	Haiku 4.5	Git, HTML/CSS/JS, Lighthouse, Low context	5	Accessibility Auditor, Responsive Tester, Perf Optimizer
Hephaestus	Sonnet 4.6	Git, SSH, Docker, Monitoring, Medium context	5	Deployment Orchestrator, Health Monitor, Backup Automator

---

Implementation Priority

Phase 1 (Essential — Deploy Day 1)

• Standard tools for all agents
• Models locked in
• Git/SSH access configured

Phase 2 (High Impact — Deploy Week 1)

• Daedalus: ADR Generator + Scale Analyzer
• Talos: Code Optimizer + API Generator
• Icarus: Accessibility Auditor + Responsive Tester
• Hephaestus: Deployment Orchestrator + Health Monitor

Phase 3 (Nice to Have — Deploy Week 2-3)

• All remaining custom skills
• Integrate with monitoring/logging systems
• Optimize based on initial feedback

---

Custom Skill Development Notes

Building Custom Skills

These custom skills could be:

• OpenClaw Skills (if we build them as reusable skills)
• Agent-specific prompts (simpler, embedded in agent system prompt)
• Integration with existing tools (Lighthouse plugin, PHPStan wrapper, etc.)

Recommendation: Start with agent-specific prompts (built into system prompt), then graduate to OpenClaw Skills if they prove valuable.

Examples of Skill Prompts

For Talos (Code Optimizer prompt snippet):

When reviewing PHP code, analyze for:
- N+1 query patterns (suggest eager loading)
- Unoptimized loops (suggest collection methods)
- Missing indexes (suggest database optimization)
- Object allocation in loops (refactor out)
Report findings as: [Issue] → [Why it matters] → [Suggested fix]

For Icarus (Accessibility Auditor snippet):

When building UI, verify:
- All images have alt text
- All inputs have associated labels
- Color contrast ≥ 4.5:1 for AA compliance
- Keyboard navigation works
- Semantic HTML (no div soup)
Report: [Issue] → [WCAG criterion] → [How to fix]

---

Questions for You

1. Custom Skills: Should I build these as OpenClaw Skills or embed them in agent prompts?
2. Priority: Approve all 4 high-impact skills for Phase 2, or be more selective?
3. Monitoring: Should Hephaestus integrate with existing monitoring systems (DataDog, etc.) or use manual checks?
4. Accessibility: Is WCAG 2.1 AA the right standard for Icarus, or should we aim for AAA?

---

Recommendation

Deploy:

• ✅ All standard tools (access established, models locked)
• ✅ Daedalus: ADR Generator + Scale Analyzer (architectural excellence)
• ✅ Talos: Code Optimizer + API Generator (dev velocity)
• ✅ Icarus: Accessibility Auditor + Responsive Tester (quality + inclusion)
• ✅ Hephaestus: Deployment Orchestrator + Health Monitor (reliability)

This gives each agent 2 force multipliers immediately while keeping setup manageable.

---

Ready to implement upon your approval.