Repository navigation
Developer Credits Request: Production-Grade Copilot Agent SDK #321
Description
The GitHub Copilot CLI has powerful permission features (--allow-all-tools, --allow-all-paths) but lacks a production-grade programmatic SDK. This limits enterprise CI/CD, agent orchestration, and production deployments.
Goal: Build a production-grade SDK that surpasses Claude Agent SDK in technical capabilities.
Key Technical Differentiators
1. Enterprise-Grade Permission System
# Advanced permission management with deny precedence
config = PermissionConfig(
allow_all_tools=True,
denied_tools=['shell(rm)', 'shell(sudo)'], # Security-first
auto_allow_system_paths=True
)
# Temporary permissions with automatic reversion
async with agent.temporary_permissions(allow_all_tools=True):
await agent.execute("risky operation")vs Claude SDK: Basic allowlist only, no deny rules.
2. Robust Workflow Engine
workflow = CopilotWorkflow("Deploy")
workflow.add_step("test", "run tests")
workflow.add_step("deploy", "deploy", on_failure="rollback")
results = await workflow.run(autonomous=True, on_error='rollback', max_retries=3)vs Claude SDK: No workflow engine.
3. Production Observability
agent = CopilotAgent(
metrics=PrometheusMetrics(port=9090),
tracer=OpenTelemetryTracer()
)
# Automatic metrics: requests_total, duration_seconds, errors_totalvs Claude SDK: No monitoring features.
4. Testing Framework
class TestDeployment(AgentTestCase):
async def test_workflow(self):
with mock_tool('shell', return_value='success'):
result = await agent.execute("deploy")
self.assert_no_errors(result)vs Claude SDK: No testing utilities.
5. Performance & Security
- Connection pooling, caching, parallel execution
- Audit logging, sandbox isolation, policy enforcement
- Secret management (Vault), compliance features
vs Claude SDK: Basic security only.
Feature Comparison
| Feature | Claude SDK | Copilot Agent SDK |
|---|---|---|
| Permission System | Basic | ⭐ Deny rules + temp perms |
| Workflow Engine | ❌ None | ⭐ Rollback + retry |
| Agent Templates | ❌ None | ⭐ 5+ specialized |
| Testing Framework | ❌ None | ⭐ Mock + replay |
| Monitoring | ❌ None | ⭐ Prometheus + OpenTelemetry |
| Performance | Basic | ⭐ Pooling + caching |
| Security | Basic | ⭐ Audit + sandbox |
Current Investment & Request
Already Invested (80 hours, personal Copilot Pro+ quota - $39/month):
- ✅ Discovered 5 undocumented permission flags
- ✅ 10 comprehensive flag tests
- ✅ 700+ line tutorial
- ✅ 772+ line permissions proposal
- ✅ SDK foundation (permissions, types, structure)
- ✅ Filed Issue Comprehensive Permissions System Improvements Proposal #307 (now in Epic Epic: Permissions Improvements #316)
Request: Additional Developer Credits
I currently have GitHub Copilot Pro+ ($39/month) but need additional credits for intensive SDK development:
Usage Breakdown (~1500 premium requests/month × 4 months):
- Core Development (40%): Permission testing, process management, error handling
- Testing & Validation (30%): MCP integration, memory system, performance benchmarks
- Documentation (20%): Production examples, tutorials, API reference
- Community Support (10%): Bug fixes, code review, optimization
Why Additional Credits:
- Personal Pro+ quota is insufficient for intensive development
- Testing requires hundreds of iterations per feature
- Documentation needs validated code examples
- Enterprise features require extensive validation
Timeline & Deliverables
Month 1: v0.1.0 alpha (core agent, tests)
Month 2: v0.3.0 beta (workflows, MCP)
Month 3: v0.8.0 RC (monitoring, templates)
Month 4: v1.0.0 stable (PyPI, docs)
Success Metrics
- Code Quality: >80% test coverage
- Performance: <100ms SDK overhead
- Adoption: 1000+ PyPI downloads Month 1
- Better than Claude SDK: 6 unique technical advantages
Validation
William Martin (GitHub Staff, Epic #316 creator) confirmed in Issue #307:
"Right now I think you can achieve the overall goal here by passing --allow-all-tools and --allow-all-paths"
This validates: ✅ Features exist ✅ Need SDK wrapper ✅ Part of official roadmap
Open Source Commitment
License: MIT
Repository: https://github.com/diegofornalha/copilot-agent-sdk-python
Transparency: Public roadmap, issues, PRs
Documentation: English + Portuguese
Alternative Support
If additional credits aren't available:
- Technical partnership (architecture reviews, early access)
- Recognition (featured project, blog post)
- Community support (developer advocate, infrastructure)
I'm building this regardless - additional credits would accelerate from 12+ months to 4 months and enable production-grade features.
Goal: Make Copilot CLI the platform of choice for production AI agents.
Contact: @diegofornalha
Repository: https://github.com/diegofornalha/copilot-agent-sdk-python
Related: Issue #307, Epic #316
Activity
diegofornalha commentedon Oct 15, 2025
@williammartin Can you tell me who can help me with this?
diegofornalha commentedon Oct 15, 2025
Update: Community-Driven Development
I'd like to clarify and strengthen this proposal:
Organization: Agentes Integrados
GitHub: https://github.com/AgentesIntegrados/
Rather than individual credits, I propose sponsoring the @AgentesIntegrados organization with GitHub Copilot Pro+ access. This enables:
Community Collaboration:
Current Team:
Updated Request
Instead of individual credits, provide:
GitHub Copilot Pro+ for the @AgentesIntegrados organization for 1 year
Benefits:
✅ Multiple developers can contribute simultaneously
✅ Faster development through parallel work
✅ Better code review with team collaboration
✅ Sustainable maintenance beyond initial release
✅ Community building around the SDK
✅ Knowledge transfer to more developers
Deliverables (Organization-backed):
Why This Is Better
For GitHub:
For Users:
ROI Comparison
Organization sponsorship ($39/month × 12 = $468/year):
vs Alternative:
This community-driven approach ensures the SDK becomes a sustainable, well-maintained project rather than a solo effort.
Repository will be under: https://github.com/AgentesIntegrados/copilot-agent-sdk-python
Current foundation: https://github.com/diegofornalha/copilot-agent-sdk-python (will be transferred)
patniko commentedon Oct 26, 2026
Thanks for the detailed proposal and for the work you've put into this. We aren't able to provide Copilot credits or sponsorships through this issue tracker, so we're closing this request. Since this was filed, GitHub has released the official GitHub Copilot SDK, which exposes the same agent engine that powers Copilot CLI for Python, TypeScript, Go, .NET, Java, and Rust, including programmatic permission handling, hooks, and OpenTelemetry support. If you run into gaps while building on it, please open an issue in that repository.
Triaged as part of a Copilot SDK backlog review, with help from GitHub Copilot.