Add interactive agent dashboard and CLAUDE.md reference guide

[anshmudgil]

What does this PR do?

This PR introduces two major additions to the repository:

1. Interactive Agent Dashboard (docs/index.html) — A fully-featured web interface for browsing, searching, and exploring all 106+ agents in The Agency. Features include:

   • Real-time search across agent names and descriptions
   • Sidebar navigation by division with agent counts
   • Responsive grid layout with agent cards
   • Detail panel with full agent documentation, markdown rendering, and installation instructions
   • Multi-tab installation guides (Claude, Cursor, Windsurf, Aider, Gemini CLI, OpenCode, Antigravity)
   • Copy-to-clipboard and download functionality
   • Dark theme optimized for developer workflows
   • Mobile-responsive design

2. CLAUDE.md Reference Guide — A comprehensive 520-line reference document covering:

   • Repository structure and organization
   • Complete agent file format specification (YAML frontmatter + body sections)
   • Full roster of all 106+ agents organized by division with metadata tables
   • Division reference guide with counts and purposes
   • NEXUS orchestration system overview (phases, modes, principles)
   • Scripts reference (convert.sh, install.sh, lint-agents.sh)
   • Integration guide overview
   • Branch policy and contribution workflow

Why These Changes?

Dashboard: Provides an intuitive, visual entry point for discovering and understanding agents. Eliminates friction for users who want to browse the library without cloning the repo or reading markdown files. Supports all major AI coding tools with tool-specific installation instructions.

CLAUDE.md: Serves as the authoritative internal reference for developers, maintainers, and contributors. Consolidates critical information (structure, format specs, agent roster, NEXUS doctrine) in a single, machine-readable document. Enables better onboarding and reduces documentation fragmentation.

Testing

• Dashboard tested in modern browsers (Chrome, Firefox, Safari)
• Search, filtering, and panel interactions verified
• Responsive design validated at mobile (480px), tablet (768px), and desktop breakpoints
• All 106+ agents render correctly with proper metadata and colors
• Installation code blocks copy correctly to clipboard

Notes

• Dashboard data is embedded directly in HTML (no external API calls required)
• .nojekyll file added to enable GitHub Pages serving of the HTML dashboard
• Both files are additive; no existing agent files or core functionality modified

https://claude.ai/code/session_01XGvaA2yLaJi8n2n7do156b

[msitarzewski]

Hey @anshmudgil — love the dashboard concept! Making agents discoverable without cloning is a great idea.

A couple of thoughts:

1. CLAUDE.md — we already have a .claude/ project setup, so we'd want to review this for consistency. Happy to work with you on aligning it.

2. Dashboard approach — the embedded agent data in the HTML will go stale every time an agent is added or modified. We'd prefer a client-side approach: a vanilla JS page that fetches the repo's directory structure via the GitHub API, parses the .md files (frontmatter + body) at runtime, and renders everything dynamically. That way it's always up to date with zero maintenance.

   Each agent now has emoji and vibe fields in frontmatter, which would make great card displays.

Would you be interested in reworking the dashboard to use this approach? We can provide guidance on the frontmatter schema and directory structure.