Command Pattern Consistency: One Mental Model for All Workflows
How a 5-tier documentation architecture enables consistent, discoverable slash commands. Build once, reuse forever.
What if every workflow in your AI system worked exactly the same way?
Type /pentest and get a guided security assessment. Type /write and get structured content creation. Type /job-analysis and get career optimization. Same pattern every time—learn one, understand all.
That's the slash command system in Intelligence Adjacent.
The Problem: Command Chaos
Most AI workflows are ad-hoc. You explain what you want, hope the model understands, repeat context when it forgets. Every interaction feels like starting from scratch.
Think of it like a restaurant with no menu. You describe what you want to eat. The chef interprets. Sometimes you get what you expected. Often you don't.
Slash commands are the menu.
Defined workflows with clear inputs, predictable outputs, and consistent execution. The chef (agent) knows exactly what to make because you ordered from the menu.
The Solution: 5-Tier Architecture
The command system uses layered documentation—think of it like a library catalog.
Top level (README.md) tells you what commands exist, organized by category.
Command files (commands/*.md) contain everything about that specific command—usage, examples, expected outputs.
Skill files (skills/*/SKILL.md) explain the underlying methodology the command invokes.
Framework root (CLAUDE.md) provides system-level navigation following Anthropic's recommended patterns.
Public docs (blog) teach others how to build similar systems.
README.md
"Security Commands: /pentest, /vuln-scan, /code-review..."
↓
commands/pentest.md
"5-step workflow, 3 examples, agent routing"
↓
skills/security-testing/SKILL.md
"PTES methodology, tool catalog, output templates"
Each tier serves a different need. Discovery happens at the top. Execution details live deeper.
How Commands Work
Commands come in two patterns: single-phase and prompt-chained.
Single-Phase Commands
Simple workflows that collect context and execute:
1. Context Collection - Questions gather what the command needs
2. Validation - Pre-flight checks confirm prerequisites
3. Agent Routing - System selects appropriate agent and skill
4. Execution - Agent runs the workflow
5. Output - Results saved to structured directories
Most commands use this pattern: /vuln-scan, /threat-intel, /job-analysis.
Prompt-Chained Commands (Multi-Phase)
Complex workflows with explicit phases and gates:
/write uses 5 phases:
RESEARCH → DRAFT → QA → VISUALS → PUBLISH
↓ ↓ ↓ ↓ ↓
01-*.md 02-*.md 03-*.md 04-*.md 05-*.md
Key concepts:
- Phases - Discrete workflow stages with specific outputs
- Gates - Checkpoints that must pass before proceeding
- Phase Detection - System knows which phase you're in
- Metadata Tracking - JSON file tracks progress
Structure:
commands/{name}/
├── 00-WORKFLOW.md # Orchestrator
└── prompts/
├── 01-RESEARCH.md # Phase 1
├── 02-DRAFT.md # Phase 2
└── ...
Example: /write QA gate
Phase 3 (QA) requires passing a 5-point voice check before proceeding to visuals. The gate blocks progression until content meets standards.
✅ Voice Check: 5/5 passed
→ Gate PASSED
→ Ready for Phase 4: VISUALS
Commands using prompt chains:
/pentest- EXPLORE → PLAN → CODE → QA → COMMIT/code-review- ANALYZE → DOCUMENT → DELIVER/risk-assessment- INTERVIEW → RESEARCH → DELIVER/policy- IDENTIFY → ANALYZE → CUSTOMIZE → GENERATE → VALIDATE/compliance- IDENTIFY → ANALYZE → ASSESS → RECOMMEND → DELIVER/write- RESEARCH → DRAFT → QA → VISUALS → PUBLISH
This architecture enforces quality at each stage. You can't skip QA.
The Template: Build Once, Reuse Forever
Here's the key insight: I didn't build each command individually. I built one template and filled it for every command.
Reference implementation: /pentest took 4 hours to perfect—every section, every example, every edge case documented.
Template extraction: 30 minutes to replace specifics with placeholders.
New commands: ~45 minutes each. Copy template, fill sections, done.
Result: Consistent commands with zero decision fatigue. The template enforces quality.
What Commands Provide
For users:
- Predictable - Same pattern everywhere
- Discoverable - Find commands from multiple entry points
- Self-documenting - Everything explained in one file
- Guided - No need to know details upfront
For the framework:
- Orchestration - Consistent agent invocation
- File-based - Version controlled, no databases
- Scalable - Template makes infinite commands possible
Command Categories
Commands are organized into categories that map to framework capabilities:
- Security Testing — Offensive security with methodology enforcement (Director/Mentor/Demo modes)
- Advisory, Risk & Incident Response — Defensive security, risk assessments, and IR workflows
- Compliance — Regulatory framework assessments (ISO, NIST, machinery safety)
- Career & Personal Development — Job analysis, coaching, fitness programming
- Content & Infrastructure — Blog writing, diagrams, repository ingestion
- Framework Management — Internal maintenance and synchronization
For the complete current list of commands with skills and descriptions, see the IA Framework Reference: Commands.
The framework grows as new commands are added—each following the same template pattern for consistency.
Try It Yourself
# Any command starts the same way
/pentest
# Context collection begins
# "Which mode: Director, Mentor, or Demo?"
# "What's the target scope?"
# "Authorization documented?"
# "Timeline constraints?"
# "Specific areas of concern?"
# Agent executes with full context
# Results in output/engagements/pentests/{target}/
The command handles complexity. You focus on decisions.
Building Your Own Commands
Want to add commands to the system? Follow the template:
- Copy
library/templates/COMMAND-TEMPLATE.md - Fill frontmatter (name, agent, skill)
- Write 5 context collection questions
- Add 3 realistic examples
- Document expected output structure
- Save to
commands/{name}.md - Update README.md registry
Total time after practice: 30-45 minutes per command.
The Philosophy
Slash commands embody the Intelligence Adjacent principle: orchestration over intelligence.
The system's value isn't in smarter models. It's in consistent scaffolding that makes models useful. Commands provide that scaffolding—guided workflows where humans make decisions and AI handles execution.
Every command asks: What do you want to accomplish? Then it handles the how.
That's augmentation, not automation.
Related Posts
- IA Framework Architecture Overview - How commands fit the larger system
- Skills System Deep Dive - The methodology behind commands
- Agent Architecture - How commands invoke agents
Sources
Framework & Architecture
- IA Framework Repository - Full source code and command implementations
- Building Effective Agents - Anthropic - Agent design patterns
Claude Code Integration
- Claude Code Hooks - Event system for command detection
- Claude Code Skills - Skill system documentation
Methodologies Referenced
- PTES Technical Guidelines - Penetration testing methodology
- NIST Cybersecurity Framework - Security assessment standards
