Command Reference

All Agency commands start with /agency. Use natural language for quick results, or add flags for more control.

Command Overview

Command	Purpose	Speed	Control	Example
`/agency`	Natural language mode	Auto	Minimal	`/agency Build a landing page for my AI startup`
`/agency brief`	Create new brief	2 min	Guided	`/agency brief "SaaS onboarding flow"`
`/agency build`	Execute full pipeline	15-30 min	Flags	`/agency build BRIEF-001 --team`
`/agency review`	Standalone quality check	5 min	Focused	`/agency review BRIEF-001`
`/agency learn`	Collect feedback	10 min	Interactive	`/agency learn`
`/agency evolve`	Trigger skill evolution	5-15 min	Targeted	`/agency evolve --agent copywriter`
`/agency resume`	Continue interrupted build	Auto	Recovery	`/agency resume BRIEF-001`
`/agency profile`	View evolution status	1 min	View-only	`/agency profile`
`/agency sync-upstream`	Check moai-adk updates	2 min	View-only	`/agency sync-upstream`
`/agency rollback`	Revert skill generation	2 min	Targeted	`/agency rollback copywriter gen-010`
`/agency config`	View/edit settings	Instant	Admin	`/agency config`

Natural Language Mode (Just Do It)

The simplest way to use Agency: describe what you want in natural language.


/agency Build a landing page for an AI research platform 
targeting academic institutions, emphasizing data collaboration 
and team productivity

Agency will:

Ask clarifying questions about brand, audience, and success metrics
Create a BRIEF automatically
Execute the full pipeline
Deploy live website
Show URL when complete

Advantages:

Fastest path (2-3 questions, then automated)
No need to learn flags
Ideal for first-time users

Disadvantages:

Less control over specifications
Can’t review BRIEF before building
All-or-nothing execution

`/agency brief` - Create New Brief

Create a detailed project specification from your description.

Basic Usage


/agency brief "AI startup landing page for research labs"

Interactive Flow

When you submit a brief, Agency asks clarifying questions:

Question 1: Business Goals “What’s the primary success metric for this project?”

Options: Signup conversion, Trial signups, Lead generation, Brand awareness

Question 2: Target Audience “Who’s the main decision-maker visiting this site?”

Options: Researcher, Lab Manager, University Administrator, Grant Manager

Question 3: Brand Context “Do you have existing brand guidelines?”

Options: Yes (provide link), No (create basic profile), Using template

Question 4: Timeline & Budget “When do you need this live?”

Options: ASAP (2 weeks), Standard (4 weeks), Flexible (8+ weeks)

Output

Creates .agency/briefs/BRIEF-001/spec.md containing:


# BRIEF-001: AI Research Platform Landing
 
## Goal
Increase trial signups from academic institutions by 15% within 90 days
 
## Outcome
Professional landing page deployed in 2 weeks, generating 
qualified leads for B2B sales team
 
## Context
Brand: Technical, trustworthy, innovation-focused
Audience: Research directors (PhD, 40-60 years old)
Technical: Next.js, PostgreSQL, Vercel deployment
 
## Requirements
1. Hero section with value proposition
2. Feature showcase (3-5 key benefits)
3. Social proof (customer testimonials)
4. Pricing/plan options
5. Contact form and demo CTA
6. Mobile-responsive design
7. SEO optimization
 
## Acceptance Criteria
- Mobile score: 90+
- Load time: < 2 seconds
- Lighthouse: 90+ across all metrics
- Accessibility: WCAG 2.1 AA
- Conversion: Form completion rate > 5%

`/agency build` - Execute Pipeline

Run the full pipeline to transform a BRIEF into a live website.

Basic Usage


/agency build BRIEF-001

Executes: Planner → Copywriter → Designer → Builder → Evaluator → Deploy

Step-by-Step Mode

Review output after each phase before proceeding:


/agency build --step BRIEF-001

Phase 1: Requirement Expansion Planner generates detailed specifications

Review Output

Check requirements are complete
Verify acceptance criteria are testable
Approve or request modifications

Phase 2: Creative Copywriter and Designer create content and UI specs (parallel)

Review Output

Review copy alignment with brand
Check design against brand palette
Approve or request changes

Phase 3: Implementation Builder implements code with 85%+ test coverage

Review Output

Check implementation matches design
Review code structure
Verify tests are comprehensive

Phase 4: Quality Check Evaluator runs automated tests

Review Output

Check all criteria pass
Review performance metrics
Approve deployment

Team Parallel Mode

Spawn multiple agents for faster execution:


/agency build --team BRIEF-001

Copywriter and Designer work in parallel (worktree isolation). Reduces execution time by 30-40% compared to sequential.

Resume Interrupted Build

If build interrupted, resume from last completed phase:


/agency build BRIEF-001 --resume

`/agency review` - Standalone Quality Check

Run only the evaluator phase without building.


/agency review BRIEF-001

Useful for:

Testing code quality before deployment
Checking if a previous build meets criteria
Validating performance metrics
Verifying accessibility compliance

Output

Detailed quality report:


Quality Report for BRIEF-001
────────────────────────────

Design Quality:        92% ✓ (Target: 85%)
Code Quality:          88% ✓ (Target: 85%)
Performance:           94% ✓ (Target: 85%)
Accessibility:         90% ✓ (Target: 85%)

Overall Score: 91% ✓ PASS

Detailed Results:
─────────────────

Design Quality (92%)
  ✓ Brand alignment: 95%
  ✓ Visual hierarchy: 90%
  ✓ Spacing consistency: 91%
  ✓ Typography: 90%

Code Quality (88%)
  ✓ Test coverage: 87% (Target: 85%)
  ✓ Type safety: 92%
  ✓ Documentation: 84% (Warning: below target)
  ✓ Linting: 89%

Performance (94%)
  ✓ Load time: 1.2s (Target: < 2s)
  ✓ Lighthouse: 96
  ✓ Core Web Vitals: PASS
  ✓ Mobile score: 92

Accessibility (90%)
  ✓ WCAG 2.1 AA: PASS
  ✓ Keyboard navigation: PASS
  ✓ Screen reader: 89%
  ✓ Color contrast: PASS

Recommendations:
─────────────────
1. Increase documentation strings in 3 utility functions
2. Consider lazy-loading images below fold for better CLS
3. Add skip-to-content link for accessibility

`/agency learn` - Feedback Collection

Collect feedback to teach the system about what works.


/agency learn

Interactive Feedback Session

Section 1: Overall Satisfaction “On a scale of 1-10, how happy are you with the result?”

1-4: Significant issues (triggers detailed improvement flow)
5-7: Good, with specific improvements
8-10: Excellent, capture what worked

Section 2: What Worked Well “What did the system get right? Be specific.”

Examples: “The hero copy resonated with our audience”, “Love the color palette”, “Clean code structure”
System captures patterns to replicate

Section 3: What Needs Improvement “What should we change next time?”

Examples: “Too much technical jargon in copy”, “Colors don’t match our brand”, “Missing error handling”
System identifies pain points to address

Section 4: Design Feedback “How aligned is the design with your vision?”

Layout: Too sparse / Just right / Too crowded
Colors: Off brand / Close / Perfect
Typography: Hard to read / Adequate / Excellent

Section 5: Copy Feedback “How’s the messaging resonating?”

Tone: Matches brand (yes/no)
Clarity: Clear vs Confusing
CTAs: Compelling vs Generic

Section 6: Technical Feedback “Any bugs or technical issues?”

Performance: Any slow sections?
Functionality: Anything broken?
Mobile: Any layout issues?

Output

Feedback stored in .agency/feedback/BRIEF-001/feedback.md:


# Feedback for BRIEF-001
 
Overall Rating: 8/10
────────────────────
 
What Worked Well
────────────────
1. Hero copy perfectly captures our mission
2. Feature descriptions are benefit-focused (not feature-focused)
3. Color palette feels premium and trustworthy
4. Code is clean and well-documented
5. Mobile experience is flawless
 
Needs Improvement
─────────────────
1. Testimonials section feels generic (need research-specific quotes)
2. Pricing strategy unclear (should we show enterprise plan?)
3. Legal footer missing GDPR/CCPA privacy information
4. CTA button color should match brand blue, not default
 
Design Feedback
───────────────
- Layout: Perfect
- Colors: Need adjustment (one color)
- Typography: Excellent
 
Copy Feedback
─────────────
- Tone: Matches brand
- Clarity: Clear
- CTAs: Compelling
 
Technical Feedback
──────────────────
- No bugs found
- Mobile: Perfect
- Performance: Excellent
- Note: Contact form should redirect to Hubspot
 
Evolution Triggers
──────────────────
[System identifies]:
- Testimonials pattern: Need customer-specific quotes
- Legal compliance: GDPR/CCPA templates
- CTA optimization: Color choice matters

`/agency evolve` - Trigger Skill Evolution

Manually trigger learning loop to improve agents.


/agency evolve

This analyzes ALL feedback across all projects and promotes high-confidence rules.

Evolution by Agent

Evolve specific agents only:


/agency evolve --agent copywriter

Options: copywriter, designer, builder, evaluator, planner

Evolution Dry-Run

Preview what would change without applying:


/agency evolve --dry-run

Shows:

Rules ready for promotion
Anti-patterns to block
Confidence scores
Next generation version numbers

Output

Generates skill module updates:


Evolution Complete: Generation Updates
──────────────────────────────────────

Copywriter: v2.0 → v2.1
  New Rules:
    + Benefit-first messaging (confidence: 0.75)
    + Research institution persona patterns (0.72)
    + Video CTA for complex features (0.68)

Designer: v1.5 → v1.6
  New Rules:
    + Component spacing preferences (0.80)
    + Color palette customizations (0.71)
  Anti-Patterns:
    - Avoid generic stock photos (marked forbidden)

Builder: v2.0 → v2.1
  Improved:
    + Error boundary patterns (updated)
    + Database query optimization (refined)

Files Updated:
  .agency/skills/copywriter/gen-021/
  .agency/skills/designer/gen-016/
  .agency/skills/builder/gen-021/

Next Build will use new generations automatically.

`/agency resume` - Continue Interrupted Work

Resume a build that was interrupted or failed.


/agency resume BRIEF-001

Picks up from the last completed phase:


Resume Status for BRIEF-001
───────────────────────────
Last Completed Phase: Designer (Phase 2)
Next Phase: Builder (Phase 3)
Time Elapsed: 5 minutes
Time Remaining: 10-15 minutes

Resume? [yes/no]

`/agency profile` - View Evolution Status

See how Agency has learned from your projects.


/agency profile

Output


AI Agency Evolution Profile
────────────────────────────

Projects Completed: 12
Feedback Sessions: 11
Active Rules: 18

Agent Generations
─────────────────
Copywriter:  v2.3 (11 rules, 8 active)
Designer:    v1.8 (6 rules, 5 active)
Builder:     v2.1 (7 rules, 6 active)
Evaluator:   v1.5 (3 weights, all active)
Planner:     v1.0 (no custom rules)

Learning Velocity
─────────────────
Projects/Week:         2.4
Rules Promoted/Month:  1.8
Confidence Gain:       +0.18/project
Anti-Patterns Found:   2

Skill Specialization
────────────────────
Most Confident Rules:
  1. Benefit-first messaging (0.92)
  2. Team collaboration CTAs (0.88)
  3. Research institution tone (0.85)

Custom Brand Voice
──────────────────
Copywriter has learned 3 custom patterns specific to
your brand vocabulary and audience preferences.

Next Evolution
──────────────
Ready for automatic evolution after next project.
Run: /agency evolve

Upstream Status
───────────────
moai-adk: Synchronized (v3.2.0)
Conflicts: 0
Last Sync: 2 days ago
Pending Updates: 0

`/agency sync-upstream` - Check moai-adk Updates

Check for and merge updates from moai-adk core framework.


/agency sync-upstream

Output


MoAI-ADK Upstream Check
───────────────────────

Current Version: moai-adk v3.2.0
Latest Available: moai-adk v3.2.1

Available Updates
─────────────────

1. moai-lang-python
   Changes: Type hint improvements, FastAPI patterns
   Conflicts: 0
   Recommendation: SAFE (no Agency modifications)

2. moai-workflow-testing
   Changes: New test patterns, coverage improvements
   Conflicts: 0
   Recommendation: SAFE (recommended)

3. moai-library-mermaid
   Changes: New diagram types, rendering optimization
   Conflicts: 0
   Recommendation: SAFE (recommended)

Summary
───────
3 updates available
0 conflicts detected
Recommendation: Apply all updates

Apply Updates? [yes/no]

`/agency rollback` - Revert Skill Generation

Revert a skill module to a previous generation if evolution caused problems.


/agency rollback copywriter gen-010

Rollback Criteria

Rollback when:

New generation decreased quality (evaluator rejects more)
New patterns contradict brand voice
Anti-pattern incorrectly identified

Output


Rollback: Copywriter gen-011 → gen-010
─────────────────────────────────────

Current: gen-011 (broken)
  Rules: 12 (9 working, 3 causing issues)
  Confidence Avg: 0.71
  Quality Impact: -4% (down from 92% to 88%)

Target: gen-010 (stable)
  Rules: 11 (all working)
  Confidence Avg: 0.75
  Quality Impact: +2% (was 92% when current)

Changes Discarded:
  - Over-aggressive benefit messaging (gen-011 only)
  - Removed testimonial strategy (gen-011 only)

Rollback Confirmed. Copywriter now using gen-010.
Next build will use gen-010 until next evolution.

Historical Note:
  Rollback recorded for learning: Why gen-011 failed?
  Next evolution will avoid similar patterns.

`/agency config` - View/Edit Settings

View and modify Agency configuration.


/agency config

Configuration Sections

Brand Configuration


brand:
  name: "My Company"
  voice: "Professional yet approachable"
  colors:
    primary: "#0066CC"
    secondary: "#FF6B35"
  fonts:
    sans: "Inter"
    serif: "Merriweather"

Project Settings


project:
  default_model: "sonnet"  # sonnet or opus
  max_iterations: 5
  deployment_target: "vercel"
  auto_deploy: true

Evolution Settings


evolution:
  auto_promote: true        # Automatically promote rules at thresholds
  confidence_threshold: 0.7 # Minimum confidence to apply
  safety_override: false    # Require human approval for all changes

Integrations


integrations:
  github: "org/repo"        # For deploying code
  hubspot: true             # For form submissions
  analytics: "posthog"      # Analytics provider

GAN Loop Details

When evaluator rejects output (quality < 80%), learner agent analyzes failure and improves:

GAN Loop Iterations

Iteration 1: Evaluator identifies issue (e.g., “Copy too technical for audience”)

Iteration 2: Learner recommends improvement (“Simplify jargon, use analogies”)

Iteration 3: Generator (Copywriter) revises based on feedback

Iteration 4: Evaluator re-tests

Repeat until quality passes or max iterations reached

Iteration Limits

Simple projects: 3 iterations max
Standard projects: 5 iterations max (default)
Complex projects: 7 iterations max

Escalation

If max iterations exceeded without passing:

Log issue to .agency/escalations/
Send summary to user
Require manual intervention or project pause
Mark as anti-pattern for future evolution

Configuration Reference

Key configuration locations:

Path	Purpose	Frequency
`.agency/config.yaml`	Main settings	Never (modify with `/agency config`)
`.agency/context/`	Brand constitution	Rarely (when brand changes)
`.agency/briefs/`	Project specs	Every project
`.agency/feedback/`	Learning data	Every project
`.agency/learnings/`	Pattern database	Auto-updated
`.agency/skills/`	Agent modules	Auto-evolving

Data Locations

Location	Content	Access
`.agency/briefs/BRIEF-XXX/`	Project specifications	Read
`.agency/feedback/BRIEF-XXX/`	User feedback	Read
`.agency/learnings/`	Pattern database	View only (modified by evolution)
`.agency/skills/`	Agent skill modules	View only (auto-evolving)
`.agency/escalations/`	Failed builds & issues	Read
`.agency/generations/`	Historical skill versions	Archive
`.agency/deployed/`	Live site URLs and configs	Read

Next Steps

Ready to start? Begin with /agency brief "your project" to create your first specification, then /agency build BRIEF-001 to launch the pipeline.

Or explore the Getting Started guide for a detailed walkthrough.

Command Reference

Command Overview

Natural Language Mode (Just Do It)

/agency brief - Create New Brief

Basic Usage

Interactive Flow

Output

/agency build - Execute Pipeline

Basic Usage

Step-by-Step Mode

Team Parallel Mode

Resume Interrupted Build

/agency review - Standalone Quality Check

Output

/agency learn - Feedback Collection

Interactive Feedback Session

Output

/agency evolve - Trigger Skill Evolution

Evolution by Agent

Evolution Dry-Run

Output

/agency resume - Continue Interrupted Work

/agency profile - View Evolution Status

Output

/agency sync-upstream - Check moai-adk Updates

Output

/agency rollback - Revert Skill Generation

Rollback Criteria

Output

/agency config - View/Edit Settings

Configuration Sections

GAN Loop Details

GAN Loop Iterations

Iteration Limits

Escalation

Configuration Reference

Data Locations

Next Steps

`/agency brief` - Create New Brief

`/agency build` - Execute Pipeline

`/agency review` - Standalone Quality Check

`/agency learn` - Feedback Collection

`/agency evolve` - Trigger Skill Evolution

`/agency resume` - Continue Interrupted Work

`/agency profile` - View Evolution Status

`/agency sync-upstream` - Check moai-adk Updates

`/agency rollback` - Revert Skill Generation

`/agency config` - View/Edit Settings