swe-copilot-agents

A collection of GitHub Copilot custom agents for the complete software development lifecycle — from planning through implementation, testing, and documentation.

Installation

pip install swe-copilot-agents

Then install agents into your project:

cd /path/to/your/project
swe-copilot-agents

This copies 7 .agent.md files to .github/agents/ in your project:

✓ rug-orchestrator.agent.md
✓ brainstormer.agent.md
✓ plan-executor.agent.md
✓ plan-reviewer.agent.md
✓ pytest-agent.agent.md
✓ knowledge-graph-agent.agent.md
✓ readme-generator.agent.md

Agents

See docs/AGENTS.md for full details on each agent.

Usage

Start with the orchestrator for end-to-end workflow:

@rug-orchestrator
I want to add a user authentication system with JWT tokens

Or invoke agents directly:

@brainstormer
I need to add a search feature to my application

@plan-reviewer
Review the implementation against docs/plan/2025-03-02-auth.md

How it works

User Request
      ↓
@rug-orchestrator (Pure Delegator)
      │
      ├─ Phase 1: @brainstormer → Plan document
      │           ↓ Human Approval
      ├─ Phase 2: @plan-executor → Implementation (per task)
      ├─ Phase 3: @plan-reviewer → Validate
      │           ├─ FAIL → Retry @plan-executor (up to 3×)
      │           └─ PASS → Continue
      └─ Phase 4: @pytest-agent → Tests (optional)

Each task is validated before moving on. Failed tasks are retried up to 3 times, then escalated to the human.

Testing

The package has a cross-platform test suite covering macOS, Linux, and Windows.

pip install -e ".[dev]"
pytest tests/ -v

CI runs automatically on every push across 3 OSes × 5 Python versions (3.9 – 3.13):

Tests cover:

• install() — directory creation, file copying, idempotency, error handling
• Agent file validity — Windows-safe filenames, UTF-8 encoding, required YAML frontmatter
• Cross-platform path safety — pathlib usage, no hardcoded separators

Contributing

See CONTRIBUTING.md.

License

MIT — see LICENSE.