This repository contains my personal configuration for GitHub Copilot Agent in VS Code.
It can serve as a useful starting point for those new to GitHub Copilot in VS Code.
These instructions are designed to help achieve effective responses to zero-shot prompting during typical development tasks.
- VS Code: Version 1.111
- Required MCP commands:
npx,docker,uvx - Tested models:
Claude Sonnet 4.6
Note: Behavior may vary depending on the model. Each model interprets instructions and context differently.
.github- .github/agents: Custom Agents
- .github/instructions: Custom Instructions
- .github/skills: Agent Skills
- .github/copilot-instructions.md: Workspace-optimized instructions
.vscode- .vscode/mcp.json: Default MCP server configuration
- .vscode/settings.json: Default VS Code settings
- copilot-agent.instructions.md: Single Source of Truth — enforces 3 mandatory behaviors (sequential-thinking, skill gate, subagent delegation) on every response, with a 5-phase execution protocol and forbidden call rules
- subagent-templates.instructions.md: Reference for subagents — Context Package contract, Sequential vs. Parallel decision guide, and copy-paste prompt templates for research and implementation subagents
- investigation-mode: Stops implementation when the same failure recurs after 2+ fix attempts, switches to evidence-first root-cause analysis, and only resumes once a verified plan is in place
- minimalist-surgical-development: Keeps changes as small as possible — prefers existing utilities, avoids new abstractions, and never refactors code that wasn't part of the original request
- root-cause-tracing: Traces a bug backward through the call stack or data propagation chain to find where it actually originates, rather than patching where it surfaces
- task-direction-approval: When a change in direction is needed (different library, new dependency, architectural shift, or alternative approach), explains the root cause, offers 2–3 options with trade-offs, and waits for explicit user approval before proceeding
- uncertainty-verification: Looks up exact command flags, config keys, API paths, and version-specific behavior in official documentation before stating them — never relies on assumed or remembered specifics
- verification-before-completion: Requires running the relevant verification commands and confirming their output before claiming that any work is complete or fixed — evidence first, always
- critical-thinking.agent.md: Question assumptions and guide toward optimal solutions
- mentor.agent.md: Support engineer growth through critical questioning and mentorship