The Agency: 283 AI agents, how they work, and what I use

1Mission Grokcos

In the README's words, The Agency is "a complete AI agency at your fingertips": a growing collection of agent personalities that was "born from a Reddit thread and months of iteration". Each agent is meant to be:

• Specialised: deep expertise in one domain, not a generic prompt template.
• Personality-driven: a distinct voice, communication style and approach.
• Deliverable-focused: real code, processes and measurable outcomes.
• Production-ready: tested workflows and success metrics.

The design philosophy adds two more traits: proven step-by-step workflows, and a "learning memory" section so an agent improves with use. The agents are installed into tools such as Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot, Windsurf and Aider, through the Agency Agents desktop app or the scripts in scripts/, or read and adapted by hand. Source: README.md.

3How it works Grokcos

The agent file layout

CONTRIBUTING.md sets one structure for every agent. On top is a YAML frontmatter block; below it come fixed Markdown sections.

---
name: Agent Name
description: One-line description of the specialty and focus
color: colorname or "#hexcode"
emoji: 🎯
vibe: One-line personality hook
services:            # optional, only if external services are needed
---
# Agent Name
## 🧠 Your Identity & Memory      role, personality, memory, experience
## 🎯 Your Core Mission            primary responsibilities and deliverables
## 🚨 Critical Rules You Must Follow
## 📋 Your Technical Deliverables  code samples, templates, frameworks
## 🔄 Your Workflow Process        discovery → planning → execution → review
## 💭 Your Communication Style
## 🔄 Learning & Memory
## 🎯 Your Success Metrics          quantitative and qualitative
## 🚀 Advanced Capabilities

The contributing guide groups these into a persona (who the agent is: identity, communication style, critical rules) and operations (what it does: mission, deliverables, workflow, metrics). divisions.json is the source of truth for the division list, and a lint script checks frontmatter and colours.

Handoffs and approvals

Coordination lives in strategy/, under the name NEXUS. It turns separate agents into a pipeline with defined handoffs and sign-offs:

• Handoff templates. strategy/coordination/handoff-templates.md has seven: standard handoff (metadata, context, deliverable request, quality expectations), QA pass, QA fail with retry instructions, escalation report, phase-gate handoff, sprint handoff and incident handoff.
• Dev ↔ QA loop. Work goes back and forth between a builder and a tester. After a maximum of three retries on one task, it escalates with an escalation report instead of looping forever.
• Quality gates (approval checkpoints). strategy/nexus-strategy.md puts a gate between every phase with a named gate keeper. For example, the Phase 1 gate needs a dual sign-off from the Studio Producer and the Reality Checker, and the Reality Checker is the sole authority on the final Phase 4 gate. It defaults to "needs work" until there is strong evidence.
• Activation prompts. strategy/coordination/agent-activation-prompts.md holds ready prompts for spinning up each role in a pipeline, led by the Agents Orchestrator.

Strategy, playbooks and runbooks

Install and convert scripts (described, not run)

• scripts/convert.sh converts the agent Markdown into each tool's own format (Cursor rules, Gemini CLI agents, Codex TOML, Aider and Windsurf roster files, and others) and writes them to integrations/.
• scripts/install.sh copies those converted files into each tool's config folder, for example ~/.claude/agents/ for Claude Code. You can install everything or pick divisions and agents.
• scripts/lint-agents.sh and check-divisions.sh are the guard rails that run in CI/CD to keep frontmatter and the division list consistent.

4What I use from it Grokcos

Not adopted for now: Code Reviewer, because most of my pages are single files with no second author. It is worth revisiting for a repo with real code and tests.

5Credit and licence

All agents, strategy docs and scripts are the work of the msitarzewski/agency-agents project and its contributors, released under the MIT License (© 2025 AgentLand Contributors). This page only indexes and summarises them. Briefs are quoted or shortened from each file's own description, and the files themselves are the authority.

6Glossary

Every abbreviation on this page is underlined with dots. Hover over it, tap it or tab to it to see the full form, then use Copy to take it into your own search or notes.

Abbreviations

Copy all abbreviations

Terms