covenant-cli

Scaffold governed agent services. Exit reports, typed I/O, memory inheritance -- from the first line of code.

Why Covenant CLI?

Every AI agent framework gives you tools to build agents. None give you tools to govern them.

Covenant CLI scaffolds projects where governance is structural, not aspirational:

• Exit reports -- every service run writes what worked, what failed, and what the next run should know
• Typed I/O -- Pydantic models for every agent input and output, no Dict[str, Any]
• Memory inheritance -- agents read prior learnings before acting, not after failing
• Registry tracking -- every service and agent registered before it runs
• Convention rules -- IDE-native governance rules for Cursor and Claude Code

The difference between an agent that works once and an agent you can trust is governance.

Quick Start

pip install covenant-cli

# Create a governed project
covenant init my-project
cd my-project

# Add a governed service
covenant add-service research-agent

# Check project health
covenant status

Commands

User Journeys

Four documented paths through the tool:

• First-Time Setup -- scaffold a governed project from scratch
• Iterative Development -- the learning loop across runs
• Debugging a Failing Agent -- trace failures through exit reports
• Team Onboarding -- orient a new team member in minutes

The Governance Delta

Without covenant-cli, your agents:

• Run without purpose documentation
• Pass untyped data between steps
• Fail silently and repeat mistakes
• Have no memory of prior runs

With covenant-cli, your agents:

• Declare their mandate before acting
• Use Pydantic models for all I/O
• Write exit reports on every run
• Read inheritance before starting
• Track token usage and cost per run
• Coordinate through structured memos

Same agent SDK. Same model. Different discipline.

Project Structure

my-project/
├── GOVERNANCE.md              # 18 governance rules
├── registry/agents.json       # Service and agent registry
├── memory/
│   ├── inheritance/           # Exit reports from prior runs
│   └── memos/                 # Cross-service communication
├── .cursor/rules/             # Cursor IDE governance rules
├── .claude/rules/             # Claude Code governance rules
└── services/
    └── my-service/
        ├── manager.py         # Orchestrator with exit reports
        ├── agents/            # One agent per file
        ├── schemas/types.py   # Pydantic models
        └── tools.py           # Shared tools

What's New in v0.5.0

• covenant doctor -- validate project environment health: Python version, SDK, API keys, registry integrity, governance, memory directories, IDE rules, and service scaffolding

v0.4.0

• Usage tracking -- token consumption and cost per service run, visible in covenant status
• covenant memo -- cross-service communication via structured memos (send/list/read)
• covenant consolidate -- distill exit reports into summaries with optional archiving
• covenant remember -- now searches memos and consolidated summaries (compiled-truth boost)
• Status dashboard -- unread memo count, last consolidation date, token totals per service
• 3 new governance rules -- track usage, communicate via memos, consolidate regularly

Credits

Inspired by mutta by Alejandro Garcia Polo.

License

Covenant Public License v1.0 -- free for individuals and teams under $1M revenue. Contact the author for commercial licensing.