claude-code-slash-commands-playbook

Claude Code Slash Commands Playbook — Solanasis LLC

Date: 2026-03-25 Version: 1.1 Status: AUTHORITATIVE — Single reference for all Claude Code commands, skills, and agent workflows Author: Claude Code (Opus 4.6), commissioned by Dmitri Zasage

---

TL;DR — The single “get shit done” command: /ultra-auto {description} Plans it, builds it, tests it, reviews it, verifies it. Walk away and come back to results. See Section 2c for details. Updated 2026-03-25.

---

How to Use This Playbook

This is the “what command do I use for X?” reference. It covers every slash command, skill, and agent workflow across all of Dmitri’s repos and tools. Start with the Decision Tree (Section 1), then drill into the relevant system.

Related docs:

• Ultimate Claude Code Skills Playbook — architecture, plugins, installation
• Model Council Playbook — multi-model orchestration theory
• Cheatsheet — quick-reference card for planning skills

---

1. Quick Reference: Decision Tree

What are you trying to do?
|
+-- Simple task (< 15 min, routine)
|   +-- Just do it. No planning command needed.
|   +-- Examples: research a prospect, draft an email, update a doc
|
+-- Quick research / lookup
|   +-- Simple factual lookup --> WebSearch directly
|   +-- Deep prospect/market research --> research-agent subagent
|
+-- Medium task (15-60 min, 1-2 system areas)
|   +-- /deep-plan --> implement manually using plan checklist
|   +-- Requirements unclear? --> /clarify first
|
+-- Complex task (1+ hours, 3+ system areas)
|   +-- Want full autonomy?
|   |   +-- /ultra-auto {ticket or description}
|   +-- Want to review plan first?
|   |   +-- /ultra-plan {ticket} --> review --> /ultra-execute
|   +-- Matchkeyz project?
|   |   +-- /a-planning-start --> /my-execute-plan
|   +-- GSD project?
|       +-- /gsd:new-project --> /gsd:plan-phase N --> /gsd:execute-phase N
|
+-- Validate or strengthen a plan
|   +-- /deep-plan-validate (5 Whys, trigger words, assumptions)
|   +-- /plan-review (adversarial 12-point rubric, forked context)
|
+-- Debugging
|   +-- Matchkeyz --> /my-troubleshoot
|   +-- Any project --> /deep-plan-validate (root cause via 5 Whys)
|
+-- Code review
|   +-- Matchkeyz --> /my-review-work (parallel test + review)
|   +-- Any project --> senior-reviewer subagent
|
+-- Client deliverable
|   +-- Draft it --> then ALWAYS run senior-reviewer before sending
|
+-- Domain-specific task
|   +-- Sales --> /account-research, /call-prep, /draft-outreach, etc.
|   +-- Marketing --> /campaign-plan, /content-creation, /seo-audit, etc.
|   +-- Legal --> /contract-review, /compliance-check, /triage-nda, etc.
|   +-- Compliance --> /soc-2, /hipaa, /gdpr, /iso-27001, /fedramp
|   +-- Engineering --> /architecture, /code-review, /debug, /testing-strategy
|   +-- Operations --> /runbook, /change-request, /vendor-review

---

2. The Ultra System (Flagship)

The Ultra system is the most rigorous planning and execution pipeline. It produces Sonnet-executable XML task specifications through multiple research and review cycles, then implements them through a wave-based multi-agent pipeline.

Source: c:/_my/claude-code-config/skills/ultra-*/ Deployed to: ~/.claude/skills/ via deploy.py

2a. /ultra-plan — Deep Planning (10 Phases)

Purpose: Transform a ticket, spec, or description into a comprehensive implementation plan.

Phase	Name	Actor	Context Cost
0	Input Resolution	Orchestrator	~2%
1	Clarification Gate	Orchestrator + /clarify	~5%
2	Research	Subagent: research-agent	~2%
3	Deep Planning	Subagent: ultra-planner (Opus)	~2%
4	Review Cycle 1	Subagent: ultra-plan-reviewer (Opus)	~2%
5	Revision (conditional)	Subagent: ultra-planner	~1%
6	Review Cycle 2	Subagent: ultra-plan-reviewer	~2%
7	Validation	Subagent: ultra-planner	~2%
8	Final Quality Gate	Orchestrator	~5%
9	Handoff	Orchestrator	~3%

Input formats: Ticket ID (e.g., SOL-42), file path (.md/.txt/.yaml), or inline description.

Artifacts produced:

• .ultra-plans/{slug}-STATE.md — session state (survives context loss)
• .ultra-plans/{slug}-RESEARCH.md — DRY analysis, existing components, blockers
• .ultra-plans/{slug}-PLAN.md — full plan with XML task specs
• .ultra-plans/{slug}-REVIEW-{N}.md — review findings per cycle

Key features:

• Session resume: if STATE file exists, continues from last phase
• Max 3 review cycles before escalating to user
• XML task format with <task>, <action>, <acceptance_criteria>, <verify>
• Wave-based parallelization: tasks in same wave are independent
• All assumptions tracked with validation methods
• Trigger word scanning: “disable”, “skip”, “workaround”, “temporary”, “ignore”, “suppress”, “hardcode”, “just”, “as needed”, “align with”, etc.

Usage:

/ultra-plan Add CSV export to reporting dashboard
/ultra-plan SOL-42
/ultra-plan path/to/requirements.md

2b. /ultra-execute — Execution Pipeline (6 Phases)

Purpose: Take an approved plan and implement it through multi-agent pipeline.

Phase	Name	Actor	Context Cost
0	Input Resolution + Pre-Flight	Orchestrator	~5%
1	Wave-Based Implementation	Subagent: ultra-executor (Sonnet) per wave	~2%/wave
2	Parallel Test + Review	ultra-test-runner + ultra-code-reviewer (parallel)	~3%
3	Gate Check	Orchestrator	~2%
4	Completion Verification	Subagent: ultra-completion-gate (Opus)	~3%
5	Wrap-Up	Orchestrator	~3%

Gate decision matrix:

Test Runner	Code Reviewer	Action
CLEAN	APPROVE	PASS to completion
CLEAN	CONDITIONAL	PASS (note P1s)
CLEAN	NEEDS WORK	RE-IMPLEMENT (P0 fixes)
REGRESSIONS	any	RE-IMPLEMENT (fix regressions first)
BUILD_FAILED	any	RE-IMPLEMENT (fix build first)

Key features:

• Captures test baseline before implementation
• Max 3 review cycles before escalating
• Completion gate independently verifies ALL requirements with evidence
• Minor gaps (1-2 requirements): auto-fix inline
• Major gaps (3+): return to implementation phase

Usage:

/ultra-execute csv-export-reporting-dashboard
/ultra-execute path/to/.ultra-plans/slug-PLAN.md

2c. /ultra-auto — Full Autonomous Pipeline

Purpose: Chain /ultra-plan then /ultra-execute with minimal human intervention.

Auto-proceed on:

• Wave completion
• Test runner CLEAN + Code reviewer APPROVE/CONDITIONAL
• Gate check PASS
• Minor completion gaps (1-2 requirements)
• Deviation Rules 1-3 (bugs, missing features, blockers)

STOP on (requires human):

• Ambiguous requirements (presents clarifying questions)
• Architectural changes (Deviation Rule 4)
• Human-action checkpoints (2FA, auth gates)
• Max review cycles exceeded (3 cycles)
• Major completion gaps (3+ requirements)

Usage:

/ultra-auto Add CSV export to reporting dashboard
/ultra-auto SOL-42

---

3. Planning Commands Comparison

Feature	/ultra-plan	/deep-plan	/a-planning-start	/gsd:plan-phase
Scope	All projects	All projects	Matchkeyz only	GSD projects
Phases	10	7	4 (start/dig/resume/complete)	Research+plan+verify
Review cycles	2 (max 3)	0 built-in	0 built-in	1
Adversarial review	Yes (ultra-plan-reviewer)	Manual via /plan-review	Manual	Built-in
Subagents used	3 (research, planner, reviewer)	1 (planner)	0	Varies
XML task specs	Yes	No (checklists)	No	No
Wave parallelization	Yes	No	No	Yes
Session resume	STATE.md file	deep-plan-validate	Planning memory	.planning/STATE.md
Token cost	High (~8 agent spawns)	Medium (~1-2 agents)	Low	Medium
Best for	Complex multi-system work	Medium complexity	Matchkeyz tickets	GSD-managed projects
Model	Opus orchestrator	Opus (planner agent)	Session model	Session model

When to use which:

• Simple (< 15 min): No planning command. Just do it.
• Medium (15-60 min, 1-2 areas): /deep-plan
• Complex (1+ hrs, 3+ areas): /ultra-plan
• Matchkeyz ticket: /a-planning-start (has Linear integration)
• GSD project with existing phases: /gsd:plan-phase N

---

4. Execution Commands Comparison

Feature	/ultra-execute	/my-execute-plan	/gsd:execute-phase
Scope	All projects	Matchkeyz only	GSD projects
Implementation agent	ultra-executor (Sonnet)	Session-level agents	Session model
Parallel test+review	Yes (2 agents)	Yes (/my-review-work)	Post-phase verify
Gate check	Automated matrix	Manual	/gsd:verify-phase
Completion verification	ultra-completion-gate (Opus)	/my-double-check or /my-triple-check	/gsd:verify-work
Max review cycles	3	Not enforced	Not enforced
Wave-based	Yes	Yes	Yes
Token cost	High (5+ agents)	Medium	Medium
Autonomous option	/ultra-auto	No	/gsd:autonomous

---

5. Agent Architecture

Core Agents (All Projects)

Deployed from c:/_my/claude-code-config/agents/ to ~/.claude/agents/.

Agent	Model	Max Turns	Role	Tools Allowed	Tools Blocked
planner	Opus	15	Planning, file-based output	Read, Grep, Glob, WebSearch, WebFetch, Write	Edit, Bash, Agent
research-agent	Opus	15	Deep research, multi-source	WebSearch, WebFetch, Read, Grep, Glob, Bash	Write, Edit, Agent
senior-reviewer	Opus	10	Quality verification	Read, Grep, Glob, Bash, WebSearch, WebFetch	Write, Edit, Agent

Ultra System Agents

Agent	Model	Max Turns	Role	Key Restriction
ultra-planner	Opus	20	Plan creation, XML task specs	No Edit, no Bash
ultra-plan-reviewer	Opus	10	Adversarial 15-point review	READ-ONLY (no Write/Edit)
ultra-executor	Sonnet	25	Implementation	Full access (Read/Write/Edit/Bash)
ultra-test-runner	Sonnet	10	Test suite execution	READ-ONLY + Bash
ultra-code-reviewer	Sonnet	10	6-dimension code review	READ-ONLY, >80% confidence only
ultra-completion-gate	Opus	10	Evidence-based requirement verification	READ-ONLY, zero tolerance

Key design principle: Agents that CREATE (planner, executor) have write access. Agents that REVIEW (reviewer, test-runner, gate) are READ-ONLY. This prevents reviewers from “fixing” issues instead of reporting them.

---

6. Context Window Strategy

The Ultra system is specifically designed to stay within Claude’s 200k context window across multi-hour workflows.

Budget Allocation

Component	Target	Max
Orchestrator state	~15%	25%
Subagent context (each)	Full 200k	—
Skill descriptions (total)	~2% (~16k chars)	—

Rules (Non-Negotiable)

1. Pass PATHS to agents, never CONTENT. Agents read files themselves with fresh 200k context.
2. After each agent: read ONLY verdict/status/metadata. First 20-30 lines. Extract 3-5 routing data points.
3. Never accumulate agent output. Each phase is self-contained. Later agents read from disk, not orchestrator memory.
4. STATE file is single source of truth. Write after every phase transition. New sessions resume from STATE.
5. Budget enforcement. If orchestrator context grows past 15% between phases, rules 1-3 are being violated.

Session Resume

Both /ultra-plan and /ultra-execute write state files to disk. If the context window runs out or the session is interrupted:

1. Start a new session
2. Re-invoke the same command with the same slug
3. It reads the STATE file and resumes from the next incomplete phase

---

7. Model Allocation

Task Type	Model	Reasoning
Planning, review, final gate	Opus	Needs deep reasoning, adversarial thinking
Implementation, testing	Sonnet	Faster, cheaper, follows precise instructions
Orchestration	Session model (usually Opus)	Coordinates agents, makes routing decisions
Domain skills (sales, marketing, etc.)	Session model	Generally lightweight

Cost Hierarchy (Approximate)

Workflow	Agent Spawns	Estimated Token Usage
/ultra-auto (full pipeline)	8-12	500k-1M+ tokens
/ultra-plan only	4-6	200k-400k tokens
/ultra-execute only	4-6	200k-500k tokens
/deep-plan	1-2	50k-150k tokens
/clarify	0	5k-20k tokens
/plan-review	0 (forked context)	30k-80k tokens
Domain skill	0	10k-50k tokens

Pro Tip: For medium-complexity work, /deep-plan then manual implementation is 3-5x cheaper than /ultra-auto. Reserve the Ultra system for work where the review cycles and gate checks genuinely prevent costly rework.

---

8. Claude Code Config System (All Projects)

Skills (Globally Deployed)

Source: c:/_my/claude-code-config/skills/ Deployed to: ~/.claude/skills/ via python deploy.py

Skill	Invoke	Triggers Automatically	Purpose
/ultra-plan	Manual	No	10-phase deep planning
/ultra-execute	Manual	No	6-phase execution pipeline
/ultra-auto	Manual	No	Autonomous plan+execute chain
/deep-plan	Manual or auto	Yes (complex tasks, 3+ areas)	7-phase planning doc
/deep-plan-validate	Manual or auto	Yes (plan validation, resuming)	5 Whys, assumption validation
/clarify	Manual or auto	Yes (ambiguous requirements)	3-7 multiple-choice questions
/plan-review	Manual only	No (disable-model-invocation: true)	Adversarial 12-point review

Solanasis Service Skills (Globally Deployed)

Skill	Invoke	Purpose
/contract-review	Manual or auto	Review contracts, NDAs, SaaS terms. CUAD-based.
/client-onboarding	Manual or auto	Discovery questionnaire, systems inventory, 30-60-90 plan
/dr-verification	Manual or auto	DR/BCP test plans, RTO/RPO verification, gap analysis
/security-assessment-report	Manual or auto	Client-ready security reports with NIST/CIS mapping

Compliance Skills (Globally Deployed)

Skill	Invoke	Framework
/soc-2	Manual or auto	SOC 2 Type I/II
/hipaa	Manual or auto	HIPAA Security/Privacy Rules
/gdpr	Manual or auto	GDPR (EU data protection)
/iso-27001	Manual or auto	ISO 27001 ISMS
/fedramp	Manual or auto	FedRAMP authorization

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 --dry-run     # Preview changes

---

9. Matchkeyz-Specific Commands

These live in c:/_my/_matchkeyz/_matchkeyz_code/qa/.claude/commands/ and are only available inside the Matchkeyz project.

Planning Commands

Command	Purpose
/a-planning-start	Initialize persistent planning memory for complex tickets
/a-planning-dig-deeper-n-confirm	Deep validation with 5 Whys, assumption checking
/a-planning-resume	Resume interrupted planning session
/a-planning-complete	Finalize planning for implementation

Execution Commands

Command	Purpose
/my-execute-plan	Multi-agent implementation pipeline
/my-plan-linear-ticket	Create implementation plan from Linear ticket
/my-new-functionality-implementation-start	Start implementing new feature
/my-pickup-implementation	Resume interrupted implementation
/my-fix-linear-ticket	Fix a specific Linear ticket
/my-fix-github-issue	Fix a specific GitHub issue

Testing and QA

Command	Purpose
/my-review-work	Test + review in parallel
/my-double-check	Secondary QA verification
/my-triple-check	Tertiary QA verification (highest rigor)
/my-extensive-qa-testing	Full QA test suite
/my-test-changes	Run tests on recent changes
/my-verify-implementation	Verify implementation matches plan
/my-verify-audit	Audit trail verification
/my-robustness-check	Edge case and error handling check

Code Quality

Command	Purpose
/my-code-review-for-coherence	Code coherence review
/my-api-security-review	API security audit
/my-fix-pattern-violations	Fix detected pattern violations
/my-check-conventions	Verify coding conventions
/my-check-cheatsheet	Quick reference for project patterns

Documentation and Utilities

Command	Purpose
/my-documentation-update	Update project docs
/my-cs-documentation-update	C# documentation update
/my-tdd-test-documentation	TDD test documentation
/my-create-handoff-doc	Create handoff document
/my-create-linear-ticket	Create a Linear ticket
/my-create-tickets-from-ai-suggestions	Bulk ticket creation
/my-troubleshoot	Structured debugging
/my-discuss-problem	Problem discussion framework
/my-find-clarifying-questions	Generate clarifying questions (Matchkeyz-specific)
/my-sql-easy	SQL query helper
/my-db-search	Database search utility
/my-db-release-qa	Database release QA

Matchkeyz Origins ⇒ Claude Code Config

Several Matchkeyz commands were generalized into portable skills:

Matchkeyz Command	Became	Key Feature Ported
/a-planning-start	/deep-plan	Multi-phase assessment, DRY research, planning memory
/a-planning-complete	/deep-plan	Quality checks, implementation checklist generation
/a-planning-dig-deeper-n-confirm	/deep-plan-validate	5 Whys, trigger words, evidence-based investigation
/a-planning-resume	/deep-plan-validate	Session resume, progress assessment
/my-find-clarifying-questions	/clarify	ULTRATHINK analysis, multiple-choice format
GSD/Codex rubric	/plan-review	12-point adversarial review, forked context

---

10. GSD Framework (v2.43.0)

Install: npm install -g gsd-pi (already installed) CLI: gsd or gsd-cli

GSD is a standalone CLI wrapping Claude Code with spec-driven, phase-based workflows.

Core Commands

Command	Purpose
/gsd:do	Smart router — dispatches to appropriate GSD command
/gsd:new-project	Initialize a new GSD project
/gsd:new-milestone	Create a new milestone within a project
/gsd:plan-phase N	Research + plan + verify for phase N
/gsd:execute-phase N	Wave-based parallel execution of phase N
/gsd:autonomous	Full chain: all phases from plan through execution

Monitoring

Command	Purpose
/gsd:progress	Show current progress
/gsd:health	Project health check
/gsd:stats	Token usage and timing statistics

Verification

Command	Purpose
/gsd:verify-work	Verify completed work
/gsd:verify-phase	Verify a specific phase

Phase Management

Command	Purpose
/gsd:add-phase	Add a new phase
/gsd:insert-phase	Insert a phase at specific position
/gsd:remove-phase	Remove a phase
/gsd:add-tests	Add tests to a phase
/gsd:add-todo	Add a TODO item
/gsd:settings	View/edit GSD settings

CLI Options

gsd                          # Interactive TUI mode
gsd -c                       # Resume most recent session
gsd -w [name]                # Start in isolated worktree
gsd --model claude-opus-4-6  # Override model
gsd headless auto            # Headless autonomous mode
gsd sessions                 # List past sessions
gsd worktree list            # List active worktrees

GSD vs Native Skills

GSD Command	Equivalent Native Skill	Notes
/gsd:plan-phase	/deep-plan	GSD adds state management
/gsd:discuss-phase	/clarify	Similar output format
/gsd:verify-work	/plan-review	GSD auto-tracks what to verify
/gsd:autonomous	/ultra-auto	Different orchestration approach
/gsd:execute-phase	/ultra-execute	GSD is phase-scoped

---

11. Knowledge Work Plugin Skills (Available but Not Auto-Loaded)

These come from the knowledge-work-plugins marketplace. They are installed but not loaded into the context budget by default. Invoke them explicitly by name.

Sales (9 skills)

Skill	Purpose
/account-research	Deep account research before outreach
/call-prep	Prepare for a sales call
/call-summary	Summarize a completed call
/competitive-intelligence	Competitive landscape analysis
/create-an-asset	Create a sales asset (one-pager, etc.)
/daily-briefing	Daily sales briefing
/draft-outreach	Draft outreach messages
/forecast	Sales forecast
/pipeline-review	Pipeline health review

Marketing (8 skills)

Skill	Purpose
/brand-review	Brand consistency review
/campaign-plan	Campaign planning
/competitive-brief	Competitive brief
/content-creation	Content creation workflow
/draft-content	Draft content pieces
/email-sequence	Email sequence design
/performance-report	Marketing performance report
/seo-audit	SEO audit

Engineering (10 skills)

Skill	Purpose
/architecture	Architecture review/design
/code-review	Structured code review
/debug	Structured debugging
/deploy-checklist	Deployment checklist
/documentation	Documentation generation
/incident-response	Incident response workflow
/standup	Standup summary
/system-design	System design document
/tech-debt	Technical debt analysis
/testing-strategy	Test strategy design

Operations (9 skills)

Skill	Purpose
/capacity-plan	Capacity planning
/change-request	Change request document
/compliance-tracking	Compliance status tracking
/process-doc	Process documentation
/process-optimization	Process optimization analysis
/risk-assessment	Risk assessment
/runbook	Operational runbook
/status-report	Status report
/vendor-review	Vendor evaluation

Legal (9 skills)

Skill	Purpose
/brief	Legal brief
/compliance-check	Compliance verification
/legal-response	Legal response drafting
/legal-risk-assessment	Legal risk assessment
/meeting-briefing	Legal meeting prep
/review-contract	Contract review
/signature-request	Signature request workflow
/triage-nda	NDA triage
/vendor-check	Vendor legal check

Other Categories

• Finance: audit-support, close-management, financial-statements, journal-entry, reconciliation, sox-testing, variance-analysis
• Human Resources: comp-analysis, draft-offer, interview-prep, onboarding, org-planning, people-report, performance-review, policy-lookup, recruiting-pipeline
• Product Management: competitive-brief, metrics-review, product-brainstorming, roadmap-update, sprint-planning, stakeholder-update, synthesize-research, write-spec
• Design: accessibility-review, design-critique, design-handoff, design-system, research-synthesis, user-research, ux-copy
• Data: analyze, build-dashboard, create-viz, data-context-extractor, explore-data, sql-queries, statistical-analysis, validate-data, write-query
• Productivity: memory-management, start, task-management, update

Total available: 109 skills across all marketplace plugins.

---

12. Superpowers Plugin Skills (Installed)

These are from the superpowers plugin (superpowers@claude-plugins-official). Installed and active.

Skill	Purpose	Invoke
/brainstorming	Explore intent, requirements, constraints before creative work	/superpowers:brainstorming
/writing-plans	Spec to structured plan	/superpowers:writing-plans
/executing-plans	Execute plans with review checkpoints	/superpowers:executing-plans
/dispatching-parallel-agents	2+ independent tasks in parallel	/superpowers:dispatching-parallel-agents
/subagent-driven-development	Independent tasks in current session	/superpowers:subagent-driven-development
/verification-before-completion	Verify work before claiming done	/superpowers:verification-before-completion
/test-driven-development	TDD workflow	/superpowers:test-driven-development
/systematic-debugging	Structured debugging	/superpowers:systematic-debugging
/using-git-worktrees	Isolated git worktrees	/superpowers:using-git-worktrees
/finishing-a-development-branch	Integration decisions	/superpowers:finishing-a-development-branch
/requesting-code-review	Request code review	/superpowers:requesting-code-review
/receiving-code-review	Receive and process code review	/superpowers:receiving-code-review
/writing-skills	Create and test new skills	/superpowers:writing-skills

Superpowers vs Ultra system:

• Superpowers = individual developer practices (brainstorming, TDD, debugging, code review)
• Ultra system = end-to-end autonomous pipeline (plan → implement → test → review → verify)
• They coexist without conflict. Superpowers skills are manual-invoke only.
• When /ultra-auto is invoked explicitly, its internal clarification replaces brainstorming.

---

13. Per-Project Customization

The Ultra system and planning skills adapt to any project via CLAUDE.md conventions.

What the Skills Read from CLAUDE.md

Convention	Where Used
Plans directory	Where to save plan files
Test commands	Pre-flight baseline, test runner
Build commands	Build verification
Lint commands	Code quality checks
Naming conventions	Task spec generation
File routing table	Where to save artifacts
Ticket integration	Input resolution (Phase 0)

Project-Specific Plan Directories

Project	Plans Directory
Solanasis docs	solanasis-docs/playbooks/plans/
Matchkeyz	matchkeyz-docs/plans/
Any other	.claude-plans/ in project root
Ultra system	.ultra-plans/ in project root

Adding Ultra to a New Project

• Ensure project has a CLAUDE.md with test/build/lint commands
• Run python c:/_my/claude-code-config/deploy.py to ensure skills are deployed
• Test with /ultra-plan {simple description} to verify setup
• Check .ultra-plans/ directory is created correctly
• Verify STATE file resume works (kill session mid-plan, restart)

---

14. Cost and Token Considerations

Cost Tiers

Tier	Commands	Est. Tokens	When to Use
Free	Just do it (no command)	5k-20k	Simple tasks
Light	/clarify, domain skills	5k-50k	Clarification, domain tasks
Medium	/deep-plan, /plan-review	50k-150k	Medium complexity planning
Heavy	/ultra-plan, /ultra-execute	200k-500k each	Complex multi-system work
Maximum	/ultra-auto	500k-1M+	Full autonomous pipeline

Daily Budget Mental Model

Assuming ~$50/day API budget:

Activity	Approx. Cost	Daily Capacity
/ultra-auto full pipeline	$5-15	3-10 per day
/ultra-plan only	$2-5	10-25 per day
/deep-plan	$0.50-2	25-100 per day
Domain skill	$0.10-0.50	100-500 per day
Simple task (no command)	$0.05-0.20	Hundreds

Pro Tip: The biggest cost driver is Opus agent spawns. /ultra-plan spawns 4-6 Opus agents. If a task genuinely needs only /deep-plan (1 Opus agent), using /ultra-plan wastes 3-5x the tokens on review cycles that add little value for simpler work.

---

15. Appendix: Complete Command Index

A. Globally Available (All Projects)

Command	Category	Auto-Triggers	Source
/ultra-plan	Planning	No	claude-code-config
/ultra-execute	Execution	No	claude-code-config
/ultra-auto	Pipeline	No	claude-code-config
/deep-plan	Planning	Yes	claude-code-config
/deep-plan-validate	Validation	Yes	claude-code-config
/clarify	Clarification	Yes	claude-code-config
/plan-review	Review	No (manual only)	claude-code-config
/contract-review	Legal	Yes	claude-legal skill
/client-onboarding	Service	Yes	Custom skill
/dr-verification	Service	Yes	Custom skill
/security-assessment-report	Service	Yes	Custom skill
/soc-2	Compliance	Yes	Custom skill
/hipaa	Compliance	Yes	Custom skill
/gdpr	Compliance	Yes	Custom skill
/iso-27001	Compliance	Yes	Custom skill
/fedramp	Compliance	Yes	Custom skill

B. Matchkeyz-Specific (40+ Commands)

See Section 9 for full listing. Key commands:

Command	Category
/a-planning-start	Planning
/a-planning-dig-deeper-n-confirm	Validation
/my-execute-plan	Execution
/my-review-work	Testing
/my-double-check	QA
/my-triple-check	QA
/my-troubleshoot	Debugging
/my-plan-linear-ticket	Ticket integration

C. GSD Commands

See Section 10. Key commands: /gsd:do, /gsd:new-project, /gsd:plan-phase, /gsd:execute-phase, /gsd:autonomous, /gsd:verify-work.

D. Knowledge Work Domain Skills (109 Total)

See Section 11. Available by explicit invocation. Not auto-loaded.

E. Subagents (Invoked via CLAUDE.md Rules, Not Slash Commands)

Agent	When Invoked
planner	Complex tasks (auto per CLAUDE.md)
research-agent	Deep research tasks (auto per CLAUDE.md)
senior-reviewer	Client deliverables, strategic decisions (auto per CLAUDE.md)

---

16. Trigger Word Reference

These words in plans or code are flagged by /deep-plan-validate and /ultra-plan:

Word	Red Flag
”disable”	Are we hiding a real bug?
“skip”	What are we avoiding?
“workaround”	What is the real fix?
“temporary”	Will this become permanent debt?
“ignore”	Are we masking an error?
“suppress”	What is being hidden?
“hardcode”	Why not config/database?
“just”	Minimizing complexity?
“appropriate”	Vague — what specifically?
“as needed”	When exactly?
“similar to”	Copy it or reference it?
“align with”	Align how? Be specific.
”properly”	What does proper mean here?
“correctly”	Define correct.
”ensure”	How? What check?
“make sure”	What verification step?
“handle edge cases”	Which ones? List them.
”follow best practices”	Which practices? Cite them.

---

Last updated: 2026-03-24. Source of truth for “what command do I use?” For architecture and plugin details, see ultimate-claude-code-skills-playbook.md.