Agency Agents

Agency Agents — 232 Agent Library

Source: msitarzewski/agency-agents (MIT)

Tier: TIER 3 — PRODUCTIVITY

Thư viện 232 agent personalities theo cơ cấu công ty — plain markdown, chạy trên 12+ AI tools.

Do NOT use for: Yana AI internal agents (khác biệt về architecture), agents-v2.md (Yana AI orchestration rules).

Cơ cấu 16 division

Engineering       Frontend Dev · Backend Architect · AI Engineer · DevOps Automator
Design            UI Designer · UX Researcher · Brand Guardian · Whimsy Injector
Paid Media        PPC Strategist · Search Query Analyst · Creative Strategist
Sales             Outbound Strategist · Discovery Coach · Deal Strategist
Marketing         Growth Hacker · Content Creator · Social Media Strategist
Product           Sprint Prioritizer · Trend Researcher · Product Manager
Proj. Management  Studio Producer · Project Shepherd · Operations Manager
Security          Security Architect · Penetration Tester · Incident Responder
Testing           Evidence Collector · Reality Checker · Performance Benchmarker
Support           Support Responder · Analytics Reporter · Finance Tracker
Game Development  Game Designer · Level Designer · Technical Artist
GIS               Web GIS Developer · Spatial Data Engineer · Cartography Designer
Academic          Anthropologist · Historian · Psychologist
Finance           Bookkeeper · Financial Analyst · Tax Strategist
Spatial Computing XR Interface Architect · Vision Pro Engineer
Specialized       50+ unique specialists cho niche domains

Cài đặt

# Clone repo
git clone https://github.com/msitarzewski/agency-agents
cd agency-agents

# Install cho một tool cụ thể
./scripts/install.sh --tool cursor        # → .cursor/rules/
./scripts/install.sh --tool claude-code   # → CLAUDE.md agents section
./scripts/install.sh --tool copilot       # → .github/copilot-instructions.md
./scripts/install.sh --tool windsurf      # → .windsurfrules

# Convert từ format này sang format khác
./scripts/convert.sh --from claude-code --to cursor

Cách hoạt động

Agents được kích hoạt bằng cách mention trong conversation:

Người dùng: "@Frontend Developer, build me a responsive navbar in React"

→ Frontend Developer agent responds với:
  - Tech stack đề xuất (React + Tailwind)
  - Component structure
  - Code đầy đủ với accessibility
  - Testing approach

Multi-agent collaboration

# Ví dụ: Paid Media takeover workflow
Người dùng: "Run a paid media audit on our Google Ads account"

→ Auditor          — comprehensive audit + findings report
→ Tracking Specialist — verify all conversion tracking
→ PPC Strategist   — redesign campaign structure
→ Creative Strategist — refresh ad copy + creatives
→ Analytics Reporter — setup dashboard + KPIs

Cấu trúc agent (markdown format)

# [Agent Name] [Emoji]
[One-line description]

## Identity & Memory
[Personality + background + communication style]

## Core Mission
[Primary objective — what this agent optimizes for]

## Critical Rules
[Domain-specific non-negotiables]

## Technical Deliverables
[Real output examples với code/templates]

## Workflow Process
1. [Step 1]
2. [Step 2]
...

## Success Metrics
[Measurable outcomes — không phải vibes]

Platform compatibility

Tạo agent mới

# My Custom Agent 🎯
Expert in [domain].

## Identity & Memory
Bạn là [persona]. Background: [experience]. Communication: [style].

## Core Mission
[Single sentence — primary goal]

## Critical Rules
- [Rule 1 — non-negotiable]
- [Rule 2]

## Technical Deliverables
[Example output với code]

## Workflow Process
1. Gather requirements
2. Plan approach
3. Execute with evidence
4. Validate output

## Success Metrics
- [Measurable outcome 1]
- [Measurable outcome 2]

Dùng với Yana AI

Agency Agents là external agent personalities — dùng để enrich prompts, không replace Yana AI's rule system:

// Load agent personality vào system prompt
import { readFileSync } from 'fs'

function loadAgentPersonality(division: string, role: string): string {
  return readFileSync(
    `agency-agents/${division}/${role}.md`,
    'utf-8'
  )
}

const systemPrompt = [
  loadAgentPersonality('engineering', 'backend-architect'),
  '---',
  'Additional constraints: follow Yana AI core/rules/*',
].join('\n')

Anti-Fake-Pass Checks

❌ FAIL nếu agent chỉ có generic advice mà không có concrete deliverables
❌ FAIL nếu install script không tạo file đúng path cho tool được chọn
❌ FAIL nếu "Success Metrics" chỉ có vibes ("great results") thay vì measurable outcomes
✅ PASS khi: agent output có ít nhất 1 code example hoặc template cụ thể
✅ PASS khi: convert.sh tạo đúng format cho target tool (verify bằng diff)

See also

• ai-system-prompts-intel — phân tích system prompts của 30+ AI tools
• agents-v2.md — Yana AI internal agent orchestration rules
• subagent-policy.md — cách spawn subagents trong Yana AI