Your GitHub workflows are manual. They shouldn't be. Deploy specialized AI agent swarms that review PRs, triage issues, run tests, and ship code while you sleep.
Built on Conduit UI, powered by Claude, zero DevOps required.
Stop context-switching between coding and ops work. Deploy agent kits that handle the boring parts:
- Architect - Analyzes requirements, designs system architecture, creates technical specs
- Implementer - Writes production code, implements features, refactors, optimizes performance
- Tester - Writes test suites, runs automated tests, validates code quality and coverage
- Reviewer - Reviews PRs, enforces coding standards, identifies security issues, provides feedback
Each kit is a swarm of specialized agents that collaborate autonomously.
Your team is:
- Waiting hours for PR reviews
- Manually triaging duplicate issues
- Running the same test suites over and over
- Deploying by hand because CI/CD is "too complex"
- Writing boilerplate code instead of solving problems
You're paying developers to do robot work.
# Download the binary
wget https://github.com/the-shit/agents/releases/latest/download/agents.phar
# Make it executable
chmod +x agents.phar
mv agents.phar /usr/local/bin/agents
# Or install globally
composer global require the-shit/agents# Deploy the PR reviewer agent swarm
agents deploy reviewer
# Deploy all kits at once
agents deploy architect
agents deploy implementer
agents deploy tester
agents deploy reviewer# Execute a specific agent kit
agents run reviewer
# Run with a specific task
agents run reviewer --task=review_pull_requests
# Run all pending tasks for a kit
agents run implementer# Watch agent activity in real-time
agents watch reviewer
# Output:
# ┌────────────────────┬─────────────────────────────┐
# │ Metric │ Value │
# ├────────────────────┼─────────────────────────────┤
# │ Kit │ reviewer │
# │ Status │ running │
# │ Progress │ 75% │
# │ Started │ 2025-12-10 14:30:00 │
# │ Tasks Completed │ 12 │
# │ Tasks Pending │ 4 │
# └────────────────────┴─────────────────────────────┘System design and planning agent.
Capabilities:
- Design system architecture
- Define component structure
- Plan data flows
- Create technical specifications
When to use: Starting new features, refactoring large systems, technical planning
agents deploy architect
agents run architect --task=analyze_requirementsCode implementation agent.
Capabilities:
- Write production code
- Implement features from specs
- Refactor existing code
- Optimize performance
When to use: Feature development, code refactoring, performance optimization
agents deploy implementer
agents run implementer --task=implement_featuresTesting and QA agent.
Capabilities:
- Write unit and integration tests
- Run automated test suites
- Validate code quality
- Check test coverage
When to use: Ensuring code quality, CI/CD pipelines, pre-deployment validation
agents deploy tester
agents run tester --task=run_test_suitesCode review and quality control agent.
Capabilities:
- Review pull requests automatically
- Enforce coding standards
- Identify security vulnerabilities
- Provide actionable feedback
When to use: PR automation, security audits, maintaining code quality
agents deploy reviewer
agents run reviewer --task=review_pull_requests# Deploy reviewer once
agents deploy reviewer
# Add to your CI/CD pipeline (.github/workflows/review.yml):
on: [pull_request]
jobs:
review:
runs-on: ubuntu-latest
steps:
- run: agents run reviewer --task=review_pull_requestsNow every PR gets:
- Code quality feedback
- Security vulnerability scans
- Style guide enforcement
- Actionable improvement suggestions
# Pre-commit hook (.git/hooks/pre-commit):
#!/bin/bash
agents run tester --task=run_test_suites
if [ $? -ne 0 ]; then
echo "Tests failed. Commit aborted."
exit 1
fiTests run automatically. Broken code never reaches main.
# 1. Design the feature
agents run architect --task=analyze_requirements
# 2. Implement it
agents run implementer --task=implement_features
# 3. Test it
agents run tester --task=write_unit_tests
# 4. Review it
agents run reviewer --task=check_code_quality
# Done. Ship it.# Check status of all deployed kits
for kit in architect implementer tester reviewer; do
echo "=== $kit ==="
agents watch $kit
echo ""
doneAgent kits are configured in config/agents.php:
return [
'kits' => [
'architect' => [
'description' => 'System design and architecture planning',
'capabilities' => [
'Design system architecture',
'Define component structure',
'Plan data flows',
'Create technical specifications',
],
'tasks' => [
'analyze_requirements',
'design_architecture',
'define_interfaces',
'create_diagrams',
],
],
// ... other kits
],
];Customize tasks, add new capabilities, or create entirely new agent kits.
Built on the Conduit UI ecosystem:
- conduit-ui/connector - GitHub API client
- conduit-ui/commit - Commit analysis
- conduit-ui/action - Actions automation
- conduit-ui/pr - PR management
- conduit-ui/issue - Issue tracking
Each agent kit uses these packages to interact with GitHub autonomously.
Powered by Claude for intelligent decision-making and code understanding.
- Solo developers - Get a full team of AI agents without hiring
- Startups - Ship faster without scaling DevOps headcount
- Enterprise teams - Automate repetitive workflows across dozens of repos
If you're manually doing work that could be automated, this is for you.
Q: Does this replace my team? A: No. It replaces the boring parts (PR reviews, test running, issue triage) so your team can focus on hard problems.
Q: What about code quality? A: Agent output follows your existing standards. It reads your style guides, linting configs, and existing patterns.
Q: Is this secure? A: Agents have read/write access to repos you configure. Use GitHub tokens with appropriate scopes. Audit logs track all actions.
Q: Can I customize agent behavior?
A: Yes. Edit config/agents.php to add tasks, modify capabilities, or create new kits.
Q: Does this work with private repos? A: Yes. Use a GitHub token with repo access.
- PHP 8.2+
- GitHub account with API token
- Laravel Zero (bundled)
Running agents across your organization? We provide:
- Custom agent kit development
- Multi-repo coordination
- Advanced workflow automation
- Dedicated support with SLA
Contact: THE SHIT
Scaling Humans Into Tomorrow
Other packages in the ecosystem:
- the-shit/chat - Multi-model chat application
- the-shit/core - Core framework
- the-shit/music - Spotify integration
# Clone the repo
git clone https://github.com/the-shit/agents.git
cd agents
# Install dependencies
composer install
# Run tests
./vendor/bin/pest
# Build binary
php agents app:buildMIT License - see LICENSE
Stop doing robot work. Deploy agent swarms.
agents deploy reviewer