Solanasis Docs

claude-code-skills-cheatsheet

4 min read

Claude Code Skills & Agents — Cheat Sheet

Quick reference for all Claude Code capabilities. Updated 2026-03-25. Full architecture: playbooks/ultimate-claude-code-skills-playbook.md Planning/execution details: c:/_my/claude-code-config/CHEATSHEET.md

The One Command: /ultra-auto

Want to drop a task and walk away? This is it.

/ultra-auto Add CSV export to the reporting dashboard

What happens autonomously:

  1. Clarifies ambiguous requirements (stops if needed)
  2. Deep codebase research (DRY analysis, existing patterns)
  3. 10-phase planning with XML task specifications (Opus)
  4. Adversarial plan review (2 cycles, 15-point rubric)
  5. 5 Whys validation + trigger word scanning
  6. Wave-based implementation (Sonnet executor)
  7. Parallel testing + code review
  8. Evidence-based completion verification (Opus)

Only stops for: ambiguous requirements, architectural changes, auth gates (2FA), stuck loops (3 cycles), major gaps (3+ requirements).

Cost: ~500k-1M tokens. Reserve for genuinely complex work.

Quick Decision: What Do I Use?

What are you doing?
    |
    +-- Simple task (< 15 min, routine)
    |       --> Just do it. No command needed.
    |
    +-- Medium task (15-60 min, 1-2 areas)
    |       --> /deep-plan --> implement manually
    |
    +-- Complex implementation (1+ hours, 3+ areas)
    |       +-- Want full autonomy?      --> /ultra-auto {description}
    |       +-- Want to review plan?     --> /ultra-plan {description}
    |       +-- Have approved plan?      --> /ultra-execute
    |
    +-- Complex non-code (research, docs, strategy)
    |       --> planner subagent (auto per CLAUDE.md)
    |
    +-- Debugging          --> /systematic-debugging
    +-- Code review        --> /requesting-code-review
    +-- Creative/design    --> /brainstorming first
    +-- Domain task        --> relevant domain skill (sales, legal, etc.)

Two Planning Tiers

Tier 1: Light Planning (deep-plan family)

SkillInvokeWhat It Does
deep-plan/deep-planPlanning doc to disk (requirements, DRY, assumptions, risks)
deep-plan-validate/deep-plan-validate5 Whys root cause, trigger words, assumption validation
clarify/clarify3-7 rich multiple-choice questions with recommendations
plan-review/plan-review12-point adversarial rubric, forked context. APPROVE/REVISE/BLOCK

Workflow: /clarify (if needed) ⇒ /deep-plan ⇒ /deep-plan-validate ⇒ /plan-review ⇒ implement

Cost: ~50k-150k tokens. Good for medium complexity.

Tier 2: Ultra System (maximum depth + autonomous execution)

SkillInvokeWhat It Does
ultra-plan/ultra-plan {desc}10-phase deep planning with multiple review cycles, XML tasks
ultra-execute/ultra-execute6-phase execution: implement ⇒ parallel test+review ⇒ verify
ultra-auto/ultra-auto {desc}Chains ultra-plan ⇒ ultra-execute autonomously

Workflow: /ultra-auto {desc} (or /ultra-plan ⇒ review ⇒ /ultra-execute)

Cost: ~500k-1M tokens. For genuinely complex multi-area work.

All Agents (9 Total)

Core Agents (all projects)

AgentModelTurnsJob
plannerOpus15Quick/medium planning, writes plan docs
senior-reviewerOpus10Quality gate for client deliverables + strategic decisions
research-agentOpus15Deep prospect/market/tech research

Ultra System Agents

AgentModelTurnsJob
ultra-plannerOpus20Deep planning, XML task specs, validation
ultra-plan-reviewerOpus10Adversarial 15-point review (READ-ONLY)
ultra-executorSonnet25Implements plan tasks with deviation rules
ultra-test-runnerSonnet10Runs tests, detects regressions (READ-ONLY)
ultra-code-reviewerSonnet106-dimension code review (READ-ONLY)
ultra-completion-gateOpus10Evidence-based requirement verification (READ-ONLY)

Pattern: Opus for decisions. Sonnet for execution.

Solanasis Service Skills (Always Available)

SkillInvokeTriggers Automatically
claude-legal/claude-legal”review contract”, “NDA”, “analyze agreement”
client-onboarding/client-onboarding”onboard client”, “new client”, “kickoff”
dr-verification/dr-verification”DR test”, “disaster recovery”, “BCP”
security-assessment-report/security-assessment-report”security assessment”, “vulnerability report”

Compliance Skills (Always Available)

SkillInvokeTriggers Automatically
iso-27001/iso-27001”ISO 27001”, “ISMS”, “information security audit”
soc-2/soc-2”SOC 2”, “trust services criteria”
hipaa/hipaa”HIPAA”, “PHI”, “healthcare compliance”
gdpr/gdpr”GDPR”, “data protection”, “privacy regulation”
fedramp/fedramp”FedRAMP”, “federal cloud”, “ATO”

Installed Plugins

PluginKey CommandsSource
superpowers/brainstorming, /systematic-debugging, /test-driven-development, /requesting-code-reviewclaude-plugins-official
commit-commands/commit, /commit-push-pr, /clean_goneclaude-plugins-official
frontend-design(auto-triggers on UI work)claude-plugins-official
sales/account-research, /call-prep, /draft-outreach, /pipeline-reviewknowledge-work-plugins
operations/vendor-review, /process-doc, /status-report, /runbookknowledge-work-plugins
engineering/standup, /code-review, /debug, /architecture, /incident-responseknowledge-work-plugins
marketing/content-creation, /campaign-plan, /brand-review, /seo-auditknowledge-work-plugins
legal/review-contract, /triage-nda, /compliance-checkknowledge-work-plugins

Where Things Live

~/.claude/                    YOU (works everywhere)
  ├── CLAUDE.md               Global instructions + planning decision tree
  ├── settings.json           Plugins + marketplace config
  ├── skills/                 16 skills (7 planning + 5 compliance + 4 service)
  ├── agents/                 9 agents (3 core + 6 ultra)
  └── plugins/                Installed plugin library

c:/_my/claude-code-config/    SOURCE OF TRUTH (version controlled)
  ├── CHEATSHEET.md           Planning/execution quick reference
  ├── deploy.py               Syncs agents + skills + CLAUDE.md to ~/.claude/
  ├── agents/                 9 agent definitions
  ├── skills/                 7 core skills + references
  └── global-claude-md/       Reference copy of ~/.claude/CLAUDE.md

Deploy Commands

python c:/_my/claude-code-config/deploy.py              # Deploy everything
python c:/_my/claude-code-config/deploy.py skills        # Skills only
python c:/_my/claude-code-config/deploy.py agents        # Agents only
python c:/_my/claude-code-config/deploy.py claude-md     # CLAUDE.md only
python c:/_my/claude-code-config/deploy.py --dry-run     # Preview without changing

Common Workflows

I want to…What to use
Build a complex feature autonomously/ultra-auto {description}
Plan a complex feature (review before building)/ultra-plan ⇒ review ⇒ /ultra-execute
Plan a medium-complexity change/deep-plan ⇒ implement manually
Research a prospect/account-research or /call-prep
Draft outreach email/draft-outreach
Review a contract/review-contract or /claude-legal
Security assessment for client/security-assessment-report + compliance skills
Onboard new client/client-onboarding
Test DR plan/dr-verification
Check SOC 2 gaps/soc-2
Generate commit message/commit
Build UI componentfrontend-design (auto-triggers)
Debug something/systematic-debugging
Get code reviewed/requesting-code-review

Ultra Auto: Mandatory Stop Points

These ALWAYS require your input (cannot be automated):

  1. Ambiguous requirements — needs your clarification
  2. Plan BLOCK after 3 cycles — planning is stuck
  3. Architectural changes — scope change needs approval
  4. Human-action checkpoints — 2FA, auth gates, etc.
  5. Max review cycles exceeded — implementation is stuck
  6. Major completion gaps — 3+ requirements unverified

If Something Isn’t Working

  1. Restart Claude Code — skills load at session start
  2. Re-deploy: python c:/_my/claude-code-config/deploy.py
  3. Plugin missing? — claude plugin install <name>@<marketplace>
  4. Skill not triggering? — invoke manually with /skill-name
  5. Context overloaded? — run /context to check for “Excluded skills” warnings

Graph View