Damon-GSY/agency

Agency — On-Demand Agent Loader

Load specialized AI agents from The Agency based on your current task context.

When to Use

When the user mentions wanting specialized agents, multiple agents, or says /agency. Also trigger when the user describes a complex multi-domain task that would benefit from domain-specific expertise.

Workflow

Step 1: Read the Agent Index

Read the cached index file at ~/.omc/agency-cache/agent-index.json. This contains 184 agents with their name, description, category, emoji, vibe, and file_path.

Step 2: Understand the Task Context

Analyze the user's current conversation/task to identify:

• Technical domains involved (frontend, backend, devops, mobile, etc.)
• Business domains involved (marketing, finance, sales, product, etc.)
• Specific skills needed (security, testing, architecture, design, etc.)

Step 3: Match and Rank Agents

Score each agent against the task context by matching keywords from:

• Agent name
• Agent description
• Agent category
• Agent vibe

Rank by relevance score. Return the top 5-8 most relevant agents.

Step 4: Present Matches to User

Display the matched agents as a selection table:

## Matched Agents for: [task summary]

| # | Agent | Category | Description |
|---|-------|----------|-------------|
| 1 | 🏗️ Backend Architect | engineering | Scalable system design... |
| 2 | 🔒 Security Engineer | engineering | Security-first... |
| ... | | | |

Then use AskUserQuestion to let the user select which agents to load. Options:

• Select specific agents by number
• "All" to load all matched agents
• "Search again" with different keywords

Step 5: Load Selected Agents

For each selected agent:

1. Copy the agent .md file from the local cache to ~/.claude/agents/:

   cp /tmp/agency-agents/<file_path> ~/.claude/agents/
   

   If the local cache at /tmp/agency-agents/ doesn't exist (e.g. after reboot), clone it first:

   git clone --depth 1 https://github.com/msitarzewski/agency-agents.git /tmp/agency-agents
   

2. Confirm each agent was loaded successfully.

Step 6: Activate

After loading, tell the user the agents are now available. They can be activated in conversation by name:

"Activate Backend Architect and help me design the API."

Step 7: Update Index (Optional)

If the user mentions the index might be stale, offer to rebuild:

python3 ~/.omc/agency-cache/build-index.py /tmp/agency-agents

Quick Commands

The user can also use these patterns:

• /agency — Interactive search and load
• /agency backend — Search for backend-related agents
• /agency security testing — Search for security + testing agents
• /agency update — Rebuild the index cache
• /agency list — Show all loaded agents in ~/.claude/agents/

Categories Available

| Category | Count | Examples | |----------|-------|---------| | academic | 5 | Anthropologist, Historian, Psychologist | | design | 8 | UI Designer, UX Architect, Brand Guardian | | engineering | 29 | Backend Architect, Security Engineer, Code Reviewer | | finance | 5 | Financial Analyst, Tax Strategist | | game-development | 20 | Unity Architect, Unreal World Builder | | marketing | 30 | SEO Specialist, Content Creator | | paid-media | 7 | PPC Strategist, Programmatic Buyer | | product | 5 | Product Manager, Sprint Prioritizer | | project-management | 6 | Project Shepherd, Jira Steward | | sales | 8 | Deal Strategist, Outbound Strategist | | spatial-computing | 6 | visionOS Engineer, XR Developer | | specialized | 41 | Salesforce Architect, MCP Builder, Recruiter | | support | 6 | Support Responder, Analytics Reporter | | testing | 8 | API Tester, Performance Benchmarker |

Notes

• Agents are loaded to ~/.claude/agents/ — this is global across all projects.
• If an agent file already exists, it will be overwritten.
• To remove a loaded agent, delete it from ~/.claude/agents/.
• The index cache lives at ~/.omc/agency-cache/agent-index.json.
• The local repo clone lives at /tmp/agency-agents/ (may be cleared on reboot).