Platform-agnostic AI agent for running AI workflows in CI pipelines, with MCP tool integration and multi-step execution engine.

• Multi-provider AI: Gemini, OpenAI, Claude (direct API and Vertex AI)
• MCP integration: Connect to any MCP-compatible tool server
• Multi-step execution: Token-aware execution engine with recovery
• YAML task definitions: DSPy-based task configuration
• Notifications: Slack and email notification support
• HTML reports: Customizable analysis report generation
• Extensible agents: Registry-based agent factory for custom agents

pip install cicaddy

# With Vertex AI Claude support
pip install 'cicaddy[vertex]'

# Run with environment file
cicaddy run --env-file .env

# Run with CLI arguments
cicaddy run --ai-provider gemini --agent-type task --log-level DEBUG

# Show configuration
cicaddy config show --env-file .env

# Validate configuration
cicaddy validate --env-file .env

Configure via environment variables or .env file:

# AI Provider (Gemini)
AI_PROVIDER=gemini
AI_MODEL=gemini-2.5-flash
GEMINI_API_KEY=your-key-here

# AI Provider (Claude via Vertex AI — uses Google Cloud ADC, no API key needed)
# AI_PROVIDER=anthropic-vertex
# AI_MODEL=claude-sonnet-4-6
# ANTHROPIC_VERTEX_PROJECT_ID=your-gcp-project
# CLOUD_ML_REGION=us-east5

# Agent
AGENT_TYPE=task
TASK_TYPE=scheduled_analysis

# MCP Servers (JSON array)
MCP_SERVERS_CONFIG=[]

# Notifications
SLACK_WEBHOOK_URL=https://hooks.slack.com/...

# DSPy Task File (takes precedence over AI_TASK_PROMPT)
AI_TASK_FILE=tasks/dora_report.yaml

Instead of raw prompt strings (AI_TASK_PROMPT), define structured tasks in YAML with typed inputs, expected outputs, MCP tool constraints, and reasoning strategy. Set AI_TASK_FILE to your task file path.

See examples/dora_metrics_task.yaml for a complete DORA metrics analysis task using DevLake MCP, and examples/templates/report_template.html for the HTML report template.

Key schema fields:

cicaddy discovers platform plugins automatically via Python entry_points. Plugins can register agents, CLI args, env vars, config sections, validators, and a settings loader — without modifying cicaddy itself.

1. Define plugin callables (my_plugin/plugin.py):

def register_agents():
    from cicaddy.agent.factory import AgentFactory
    from my_plugin.agent import MergeRequestAgent, detect_agent_type

    AgentFactory.register("merge_request", MergeRequestAgent)
    AgentFactory.register_detector(detect_agent_type, priority=40)

def get_cli_args():
    from cicaddy.cli.arg_mapping import ArgMapping
    return [
        ArgMapping(cli_arg="--mr-iid", env_var="CI_MERGE_REQUEST_IID",
                   help_text="Merge request IID"),
    ]

2. Register in pyproject.toml:

[project.entry-points."cicaddy.agents"]
my_platform = "my_plugin.plugin:register_agents"

[project.entry-points."cicaddy.cli_args"]
my_platform = "my_plugin.plugin:get_cli_args"

[project.entry-points."cicaddy.settings_loader"]
my_platform = "my_plugin.config:load_settings"

3. Install and run — plugins are discovered automatically:

pip install cicaddy my-cicaddy-plugin
cicaddy run --env-file .env

Available plugin groups: cicaddy.agents, cicaddy.cli_args, cicaddy.env_vars, cicaddy.config_sections, cicaddy.validators, cicaddy.settings_loader.

Apache-2.0