What is git-intelligence?

Giti is a command-line tool that helps you understand the health of your code repository. Think of it as a diagnostic assistant for your git project: it can tell you how active the repo is, which files keep getting changed over and over (a sign of instability), and where work was started but never finished. You install it with npm and run simple commands like giti pulse for a quick snapshot or giti hotspots to see the most volatile parts of your codebase. Beyond analysis, the project has a more ambitious layer: it describes itself as a "living codebase" that can maintain and evolve itself using AI agents. In practice, this means it runs recurring lifecycle cycles where it scans its own state, plans small tasks, writes code to fix issues, runs tests, and commits changes. The key constraint is safety. Every change has to pass multiple quality checks (called the "Immune System"), there's a kill switch to stop everything, cooldowns after regressions, and an optional mode where a human approves every merge before it happens. The tool also includes a "Field Observer" mode that lets it watch external repositories read-only and generate short narrative reports about what changed. This is useful if you maintain or monitor multiple projects and want a quick digest without digging through git logs yourself. The reports include a "mood" indicator (curious, attentive, alarmed, dozing) based on what changed since the last observation. The AI-generated narratives require an Anthropic API key, but there's a template fallback if you don't have one. The people who'd get the most value here are solo developers or small teams who want automation around repo hygiene and code quality, and who are curious about self-improving codebases as a concept. The analysis commands (pulse, hotspots, ghosts) are practical and standalone. The autonomous organism features are more experimental and require comfort with giving an AI agent write access to your repo, even with the guardrails in place. One notable tradeoff: the autonomous features depend on Claude (Anthropic's AI), so you need an API key and a token budget. The project includes an API spending cap and failure limits to keep things bounded, but anyone considering the self-running mode should start in supervised mode where merges require human sign-off.