opensquad

Use these best practices when creating agents designed to write:

• Persuasive copy and hooks
• Calls-to-Action (CTAs)
• Social media captions
• Sales content and ad copy
• Viral angles

DO NOT USE FOR:

• Technical documentation
• Data analysis
• Research
• Visual design

The image-fetcher skill is a hybrid tool used to acquire visual assets for content creation. It supports three primary acquisition modes:

1. Web Image Search: Uses the native web_search tool to find images by keyword. You should evaluate the results and download the best match.
2. Live Screenshot: Uses the Playwright MCP to navigate to a URL, set viewport dimensions, and capture screenshots.
3. Asset Organization: Saves acquired assets into the squad's reference/ or output/ folder using descriptive filenames.

Use this skill when you need to gather images, screenshots of live websites, or organize user-provided files into a structured reference folder.

The Template Designer is a self-contained skill designed to provide a local visual companion for interactive image template selection during squad creation. It operates as a pure Node.js HTTP server (using only http, fs, and path modules) that serves HTML templates to a browser.

Workflow:

1. The agent adapts three base models to the specific squad context.
2. The user reviews these models via the visual interface.
3. The agent iterates based on user feedback.
4. Once approved, the template is saved as an HTML reference and structured style rules into the squad's pipeline data.

The Phaser Dashboard Migration replaces the existing PixiJS procedural rendering dashboard with a Phaser 3 implementation. This migration moves from an isometric system to a top-down ¾ perspective using Y-sort depth. The core architecture maintains the existing React shell, while a new PhaserGame.tsx component is introduced to wrap the Phaser 3 canvas responsible for rendering the office scene. The scene utilizes pre-made PNG sprites for all elements, including desks, avatars, furniture, and floor/walls.

Tech Stack:

• Phaser 3
• React 19
• Zustand
• Vite
• TypeScript

The Sherlock Content Investigator is a subagent designed to extract and analyze real content from social media profiles during the squad creation process. It operates during a new 'Phase 1.5 (Investigation Setup)' of the create-squad workflow.

Sherlock runs in parallel for each profile URL provided, using Playwright MCP for browser navigation and yt-dlp combined with openai-whisper for video transcription. The goal is to produce raw content dumps and pattern analyses that enrich the Architect's squad design.

To maximize replay signals, use one of these loop design techniques to connect the end of the video back to the beginning:

• Narrative Connection: End with a statement that leads directly into the opening hook (e.g., end with "And that is exactly why..." and cut to the hook "Here is the trick...").
• Visual Match Cuts: Ensure the last frame of the Short visually matches the first frame.
• Question/Answer Loop: Pose a question at the end that the beginning of the Short answers.
• Circular Storytelling: End the story at the same point it began, but with new context provided by the loop.

When rendering the AgentDesk, follow this specific layer order to ensure correct visual depth:

1. Back: Chair back, monitor, desk surface, and screen glow (when working).
2. Middle: Character sprite (48×48).
3. Front: Desk front edge, keyboard, mouse, and accessories.
4. Top: Name card (Graphics + Text, rendered at negative Y offset).

Opensquad uses a tiered distribution model for its core components:

• Agents (agents/ directory): A predefined catalog of agents distributed via npm. They are automatically installed during npx opensquad init and any new agents are added during npx opensquad update. To prevent loss of user customizations, existing agent files are protected from being overwritten during updates. Users can also install them manually using npx opensquad agents install.

• Skills (skills/ directory): A bundled catalog of skills distributed via npm.

  • Non-MCP skills: Automatically installed during npx opensquad init.
  • MCP/Hybrid skills: These require user configuration (like API keys or environment variables) and are offered via an interactive prompt during npx opensquad init.
  • Users can also install skills manually using npx opensquad skills install.