Intent CLI: Semantic Git for AI Coding Sessions

What changed

The Intent CLI, a tool developed by dozybot001, offers a novel approach to managing context for AI coding agents. Unlike traditional Git, which tracks code changes, Intent focuses on recording the 'why' behind the code: the project's goals, verified milestones, and standing decisions. This semantic history is stored in plain JSON format within the .intent/ directory in a project, making it accessible to any agent or developer. The system operates on the principle of selective recording, where checkpoints are saved only when something meaningful is verified, ensuring the history remains lightweight. When resuming a session, a single command provides the next agent with the complete context, including the goal, verified state, boundaries, next steps, and applicable rules.

The CLI provides commands such as itt init to set up Intent in a project, itt intent create to define goals, itt snap create to record verified checkpoints with subsequent steps and blockers, and itt decision create to establish durable rules. The tool is designed to work alongside existing agent frameworks and IDEs like Claude Code and Codex, with specific skills and project hooks available for integration. For instance, users can install the CLI via a curl script for macOS/Linux or PowerShell for Windows, and then add the agent skill using npx skills add dozybot001/Intent -g.

Intent also offers Git-like operations for its semantic history, including itt log --graph for viewing revisions and branches, itt diff to see changes in goals and checkpoints, and itt switch -c and itt merge for branching and merging semantic histories. An optional web application, IntHub, allows for sharing and collaborating on project histories, with options for local hosting or using a hosted service at inthub.tenonai.cn. The project is currently in beta (version 8.x) and includes migration tools for users upgrading from older versions.

Why it matters for builders

Intent CLI directly tackles the persistent challenge of context loss in AI-assisted development. Developers often find themselves re-explaining project status, goals, and constraints to their AI agents after interruptions or when switching between sessions. Intent provides a structured, machine-readable memory that agents can reliably access, enabling them to pick up exactly where they left off. This reduces the cognitive load on the developer and allows AI agents to function more autonomously and effectively, accelerating the development cycle.

Practical impact

AI builders can integrate Intent into their workflows by installing the CLI and adding the corresponding agent skill. For projects already using AI coding assistants like Claude Code or Codex, setting up Intent involves initializing it within the project directory (itt init) and trusting the project hooks. Developers can then use itt commands to explicitly record key decisions and verified states, or rely on agent integrations to do so automatically. For those looking to share progress or collaborate on agent-driven projects, IntHub offers a platform to push, pull, and clone semantic histories. Developers should consider using itt migrate if upgrading from a previous major version of Intent.

Caveats and source limits

Intent CLI is currently in beta, indicating potential for breaking changes in future releases. While it integrates with popular AI coding tools, its effectiveness in reducing re-explaining in real-world scenarios is still under active testing. The source does not provide independent benchmark data on how much time or effort Intent saves compared to workflows without it. Users should be aware that the PyPI package name intent-cli is unrelated to this project; the correct package is intent-history. Integration with agents other than Claude Code and Codex is described as 'best effort' via skill instructions. The source also notes that while Git is optional, the .intent/ directory may contain internal reasoning and is excluded from Git by default, requiring deliberate commitment if team-wide sharing is desired.