A multi-agent orchestration system for GitHub Copilot. Provides a structured workflow (conductor → planner → implementer → reviewer → completion) with 16 specialized agents (11 core + 5 translation), reusable skills, and validation tooling.
Use as a shared configuration source across workspaces. Point VS Code at this repo and all projects inherit consistent agent behaviors, tool permissions, and lifecycle guardrails.
| Asset | Count | Location |
|---|---|---|
| Agents | 16 | .github/agents/ |
| Skills | 12 | .github/skills/ |
| Prompt templates | 22 | .github/prompts/ |
| Instruction files | 15 | instructions/ |
| Validation scripts | 6 | scripts/ |
| MCP servers | 6 | scripts/mcp/ |
Add these settings to your user or workspace settings.json:
{
"chat.useAgentsMdFile": true,
"chat.useNestedAgentsMdFiles": true,
"chat.instructionsFilesLocations": ["instructions"],
"chat.promptFilesLocations": [".github/prompts"],
"chat.agentFilesLocations": { ".github/agents": true },
"chat.agentSkillsLocations": { ".github/skills": true },
"chat.useAgentSkills": true,
"chat.useClaudeSkills": true,
"chat.agentCustomizationSkill.enabled": true,
"chat.customAgentInSubagent.enabled": true,
"chat.subagents.allowInvocationsFromSubagents": true,
"github.copilot.chat.searchSubagent.enabled": true,
"github.copilot.chat.customAgents.showOrganizationAndEnterpriseAgents": true,
"github.copilot.chat.cli.customAgents.enabled": true,
"github.copilot.chat.organizationInstructions.enabled": true,
"chat.thinking.style": "collapsed",
"chat.agent.thinking.collapsedTools": true,
"chat.agent.thinking.terminalTools": true,
"chat.tools.autoExpandFailures": true,
"chat.askQuestions.enabled": true,
"github.copilot.chat.anthropic.thinking.budgetTokens": 10000,
// Note: github.copilot.chat.anthropic.thinking.effort and responsesApiReasoningEffort are deprecated in 1.113.
// Configure thinking effort via the model picker submenu instead.
"github.copilot.chat.anthropic.toolSearchTool.enabled": true,
"github.copilot.chat.anthropic.contextEditing.enabled": true,
"github.copilot.chat.copilotMemory.enabled": true,
"github.copilot.chat.advanced.workspace.codeSearchExternalIngest.enabled": true,
"chat.viewSessions.enabled": true,
"chat.viewSessions.orientation": "sideBySide",
"chat.restoreLastPanelSession": false,
"chat.agentsControl.enabled": true,
"chat.agentsControl.clickBehavior": "cycle",
"workbench.startupEditor": "agentSessionsWelcomePage",
"github.copilot.chat.implementAgent.model": "Claude Sonnet 4.6 (copilot)",
"chat.tools.terminal.enableAutoApprove": true,
"chat.tools.terminal.autoApproveWorkspaceNpmScripts": true,
"chat.tools.terminal.preventShellHistory": true,
"terminal.integrated.enableKittyKeyboardProtocol": true,
"workbench.browser.openLocalhostLinks": true,
"simpleBrowser.useIntegratedBrowser": true,
"git.worktreeIncludeFiles": [".env.local", "token-thresholds.json"]
}pwsh -File scripts/validate-copilot-assets.ps1 -RepositoryRoot .pwsh -File scripts/init-artifacts.ps1This creates the local artifacts/ folder structure for session persistence.
The orchestrator works across five platforms. VS Code is native (no setup needed). The others require a one-time setup script.
| Platform | Setup | Guide |
|---|---|---|
| VS Code | Built-in — clone and open | VS Code Configuration |
| Visual Studio | powershell -File scripts/setup-vs-cli.ps1 -Strategy Symlink -TargetPath <project> |
Visual Studio Onboarding |
| Copilot CLI | powershell -File scripts/setup-vs-cli.ps1 -Strategy Symlink -TargetPath <project> |
Copilot CLI Onboarding |
| Claude Code | powershell -File scripts/setup-claude-code.ps1 -Mode Project -TargetPath <project> |
Claude Code Onboarding |
| Antigravity | powershell -File scripts/setup-antigravity.ps1 -Mode Project -TargetPath <project> |
Antigravity Onboarding |
macOS/Linux users: use the .sh equivalents (setup-vs-cli.sh, setup-claude-code.sh, setup-antigravity.sh). See Multi-Platform Setup Reference for full details.
The orchestrator ships four branches aligned to GitHub Copilot plan pricing. Push to enterprise and GitHub Actions automatically syncs the other three.
| Branch | Target Plan | Flagship Model | Use Case |
|---|---|---|---|
enterprise |
Enterprise | Claude Opus 4.6 (3x) | Source of truth - flagship reasoning on Enterprise-only Opus 4.6 |
pro-plus |
Pro+ / Business | Claude Opus 4.7 (7.5x) | Flagship reasoning via Opus 4.7 (available on Pro+, Business, Enterprise) |
pro |
Pro / Student | GPT-5.3-Codex (1x) | No Anthropic premium - routes to GPT-5.3-Codex + GPT-5.4 mini + Haiku 4.5 |
free |
Free | GPT-5 mini (0x) | Zero premium requests - all agents on GPT-5 mini or GPT-4.1 |
Each agent's model: field is a fallback array. VS Code picks the first model you have access to, so the published branches substitute unavailable models with plan-available equivalents.
- Develop on
enterprise- all agents run Enterprise-tier models with full capability. - Push to
enterprise- three GitHub Actions workflows trigger automatically:sync-pro-plus-branch.ymlresetspro-plusfromenterpriseand swaps Opus 4.6 -> Opus 4.7.sync-pro-branch.ymlresetsprofromenterpriseand swaps Anthropic premium/execution + GPT-5.4 to Pro-available models.sync-free-branch.ymlresetsfreefromenterpriseand swaps all paid models to GPT-5 mini / GPT-4.1.
- Switch tiers - clone or checkout the branch matching your plan:
git checkout pro-plus # Pro+ / Business git checkout pro # Pro / Student git checkout free # Free git checkout enterprise # Enterprise (source of truth)
Deep dive: Branching for Copilot Cost Optimization - explains the design choices, sync mechanism, and lessons learned.
16 specialized agents across three model tiers. Each agent declares a model: fallback array and a defaultEffort: hint (low/medium/high). The pro-plus / pro / free branches sync automatically from enterprise. See Model Tiers for details.
| Agent | Tier | Default Model (enterprise) | Effort | Purpose |
|---|---|---|---|---|
| Planner | Premium | Claude Opus 4.6 | high | Multi-phase planning |
| Conductor | Execution | Claude Sonnet 4.6 | medium | Lifecycle orchestration |
| Reviewer | Execution | Claude Sonnet 4.6 | high | Multi-mode code review (security mode pins Opus) |
| Implementer | Execution | Claude Sonnet 4.6 | medium | TDD implementation |
| Researcher | Execution | Claude Sonnet 4.6 | high | Evidence gathering |
| Ops | Execution | Claude Sonnet 4.6 | low | Issues, PRs, CI/CD |
| Test | Execution | Claude Sonnet 4.6 | medium | Test authoring |
| IaC | Execution | Claude Sonnet 4.6 | medium | Terraform/Bicep/Pulumi |
| GUI Tester | Execution | Claude Sonnet 4.6 | low | Browser automation |
| Translation Conductor | Execution | Claude Sonnet 4.6 | high | Translation orchestration |
| Translator | Execution | Claude Sonnet 4.6 | medium | File-level translation |
| Translation Analyzer | Execution | Claude Sonnet 4.6 | high | Dependency analysis |
| Translation Validator | Execution | Claude Sonnet 4.6 | high | Validation scoring |
| Docs | Fast | Claude Haiku 4.5 | medium | Documentation |
| UX | Fast | Claude Haiku 4.5 | low | UX/accessibility review |
| Translation Styler | Fast | Claude Haiku 4.5 | medium | Target language idioms |
defaultEffort is the role's baseline reasoning depth — individual prompts can override with the effort: frontmatter key.
| Path | Purpose |
|---|---|
.github/agents/ |
16 agent definitions |
.github/prompts/ |
22 prompt templates |
.github/skills/ |
12 agent skills |
instructions/ |
15 instruction files (global, workflow, compliance, language) |
scripts/ |
Validation and tooling (PowerShell 5.1) |
scripts/mcp/ |
6 MCP servers |
docs/ |
Guides, templates, and operational docs |
artifacts/ |
Local session outputs (plans, reviews, research, security) |
# Validate all assets
powershell -File scripts/validate-copilot-assets.ps1 -RepositoryRoot .
# Check prompt metadata
powershell -File scripts/add-prompt-metadata.ps1 -RepositoryRoot . -CheckOnly
# Run linting
powershell -File scripts/run-lint.ps1 -RepositoryRoot .
# Run smoke tests
powershell -File scripts/run-smoke-tests.ps1 -RepositoryRoot .
# Token budget report
powershell -File scripts/token-report.ps1 -Path .
# Initialize artifacts folder
powershell -File scripts/init-artifacts.ps1
# Run Pester tests
Invoke-Pester -Path tests -Output Detailed| Document | Purpose |
|---|---|
| AGENTS.md | Agent playbook, workflow guardrails, development environment |
| docs/guides/onboarding.md | New contributor guide |
| docs/guides/model-tiers.md | Model tier strategy, per-agent rationale, cost analysis |
| docs/guides/central-deployment.md | Org-level deployment with local artifacts |
| docs/guides/vscode-copilot-configuration.md | VS Code settings reference |
| docs/guides/background-agents-worktrees.md | Parallel execution with Git worktrees |
| docs/guides/claude-skills-migration.md | Migrating prompts to Claude skills |
| docs/guides/translation-guide.md | Code translation workflow guide |
| docs/operations.md | Monitoring, metrics, and backlog |
| docs/CHANGELOG.md | Version history |
See LICENSE for terms.