baseline
Baseline is an open-source governance layer for Claude Code, an AI-powered coding assistant. It provides a framework to enforce organizational policies and prevent the AI from taking unapproved actions during code development. Installed as a Node package via npx, Baseline integrates directly with Claude Code's tool lifecycle, adding a boundary where the AI cannot reach. The project is currently in public alpha and released under the Apache 2.0 license, with version v0.26 being the latest available. Baseline works by intercepting tool calls at the hook boundary. It uses PreToolUse and PostToolUse hooks to block, ask for permission, or allow actions before they execute. The system includes 27 hooks that guard against destructive commands, enforce consent markers, and validate workflow phase ordering. Since these hooks run outside the model's reach, Claude Code cannot bypass or override them by rewriting its own instructions. Beyond simple blocking, Baseline implements a comprehensive governance model with four strata: Genesis seed, Constitution, Implementation, and Tool Boundary. Rules are traced to named authorities and follow a strict precedence order, ensuring that higher-level policies cannot be silently overridden. Workflows are organized into 11 phases and 9 canonical tracks, with exceptions logged for audit visibility. User-typed gates, such as slash commands, are required for approving commits or pushes, and these consent tokens are unforgeable by the model. Baseline is designed for teams that want to safely leverage AI coding tools while maintaining compliance, security, and developer oversight. It is particularly useful in enterprise environments where audit trails and policy enforcement are critical. The project is actively developed by Friedbot Studio, and the documentation covers installation, governance concepts, and common workflows.
- Category
- LLM
- Pricing
- Free
Problem solved
Claude Code, as an AI coding assistant, can autonomously execute tool calls such as writing files, running shell commands, and making git operations, potentially performing unwanted or risky actions without explicit approval. Baseline solves this by providing a governance layer that intercepts these tool calls at a boundary the model cannot reach, blocking unapproved actions, requiring user consent for sensitive operations, and enforcing named rules through hooks. This prevents the AI from self-approving its own actions and ensures that all changes align with project policies.
Use cases
- Preventing unauthorized git commits and pushes by requiring user consent tokens
- Blocking destructive shell commands before they execute
- Enforcing phase-based development workflows from intake to commit
- Auditing AI-driven code changes and exceptions
- Running linting and test runners automatically after tool calls
- Configuring project-specific governance rules for Claude Code
- Implementing approval gates for writes in regulated environments
Audience
- Developers
- Enterprise
- Startups
Pros
- Enforcement happens outside the AI model's reach
- Blocks tool calls before they run, preventing unauthorized actions
- Configurable governance with named authority and precedence order
- Supports multiple workflow tracks and phases with audit exceptions
- Open source with permissive Apache 2.0 license
Cons
- Currently in public alpha, potential instability
- Requires Claude Code, limited to that ecosystem
- Configuration complexity may be steep for new users
- No graphical user interface; command-line driven
- Hooks only work within Claude Code's environment
Official website: https://baseline.friedbotstudio.com/
Canonical LambdaBase page: https://www.lambdabase.com/tools/baseline