Project: Creative Apps (GitHub Copilot) - Sidekick: Your CoWork for Complex Missions

Track

Creative Apps (GitHub Copilot)

Project Name

Sidekick: An Agent Harness for Developer Workflows

GitHub Username

prasan39

Repository URL

https://github.com/prasan39/sidekick

Project Description

Sidekick is an AI-powered developer cowork agent designed to execute real workflows — not just generate responses. Inspired by the idea that developers are the superheroes managing complex daily missions, Sidekick acts as the execution partner that handles the ground game.

Built on the GitHub Copilot SDK, Sidekick is an agent harness that orchestrates tools, skills, and sub-agents toward outcomes. Instead of waiting for explicit instructions, it determines which tools to invoke, chains actions together, and completes end-to-end tasks within a single session.

In a typical workflow, Sidekick can organize local files, generate applications from prompts, deploy projects live, perform live web research, and convert findings into structured artifacts like presentations. The system supports local project access, command execution, deployment workflows, and artifact generation — combining reasoning with real execution.

The application features a React + Vite frontend and an Express + WebSocket backend enabling streaming interactions, structured outputs, memory retrieval, and transparent tool-call visibility. A modular skill architecture allows developers to extend capabilities simply by adding new tools, making Sidekick adaptable to individual workflows.

Sidekick is a public, sanitized evolution of an experimental “CoWork” agent concept, redesigned for open collaboration and reproducibility. Inspired by Claude’s Cowork but customized with flexible model choices, memory controls, and developer-centric workflows, the project demonstrates how Copilot can evolve from an assistant into an outcome-oriented execution platform.

Sidekick doesn’t just chat — it ships.

Demo Video or Screenshots

https://youtu.be/s2aq79Jj4i8

Primary Programming Language

Python

Key Technologies Used

• GitHub Copilot SDK (agent orchestration)
• React + Vite (frontend UI)
• Node.js + Express (backend services)
• WebSockets (real-time streaming responses)
• Playwright MCP (live web browsing capability)
• Sub-agent architecture with intent inference
• Structured output policies for agent responses
• Vercel Preview Deploy integrations
• Presentation generation (PPTX automation)
• Retrieval + memory endpoints
• Modular skill system with dynamic loading

Submission Type

Individual

Team Members

No response

Submission Requirements

• My project meets the track-specific challenge requirements
• My repository includes a comprehensive README.md with setup instructions
• My code does not contain hardcoded API keys or secrets
• I have included demo materials (video or screenshots)
• My project is my own work with proper attribution for any third-party code
• I agree to the Code of Conduct
• I have read and agree to the Disclaimer
• My submission does NOT contain any confidential, proprietary, or sensitive information
• I confirm I have the rights to submit this content and grant the necessary licenses

Quick Setup Summary

1. Clone the repository:
   git clone https://github.com/prasan39/sidekick.git

2. Install dependencies:
   npm install

3. Configure environment variables:

   • Add GitHub Copilot token
   • Enable local auth bypass for development

4. Start the app:
   npm run dev

5. Open:
   http://localhost:5173/

Sidekick launches a local AI agent capable of interacting with project files, executing skills, and orchestrating developer workflows through a chat interface.

Technical Highlights

• Agent-first harness architecture built on GitHub Copilot SDK rather than traditional chatbot patterns.
• Intent-driven sub-agent selection using semantic inference (pptx-agent, finance-agent, live-news-agent, deploy-agent).
• Modular skill directories dynamically loaded at runtime.
• Real-time streaming over WebSockets including tool-call lifecycle visibility.
• Structured output policy enforcing predictable agent responses (answer/steps/options formats).
• Optional reasoning stream rendering for transparency.
• Playwright MCP integration enabling live browsing and real-time information retrieval.
• Automated Vercel preview deployment workflows triggered via agent actions.
• Separation between public-safe export and private memory layers for secure open sourcing.

Challenges & Learnings

One of the main challenges was moving from a chat-based assistant model to an execution-oriented agent. Designing safe local file access, command execution, and skill orchestration required careful boundaries between autonomy and developer control.

Another challenge was managing agent complexity without creating “agent sprawl.” This was addressed through intent-based sub-agent inference rather than manual routing logic.

Key learnings include:

• Developer trust increases when agents expose execution status transparently.
• Structured outputs dramatically improve usability compared to free-form responses.
• Modular skills allow rapid experimentation without rewriting core orchestration.
• AI assistants become significantly more valuable when embedded directly into local workflows rather than isolated chat environments.

The project reinforced the idea that the future of developer tooling lies in collaborative agent systems that execute alongside humans instead of merely advising them.

Contact Information

pulsar.pras@gmail.com

Country/Region

India

This issue is a project submission rather than a defined repository change, and it names no source files or tests. Start with the linked repository's README.md and the listed npm install and npm run dev entry points. There is no contributor task or completion criterion stated here.