wtf is claude-code-starter-pack?

claude-code-starter-pack is a skill for Claude Code, Anthropic's coding assistant, that sets up a specific folder of documentation files in your project. The goal is to keep a long running AI agent consistent across many separate coding sessions, so it does not quietly re-decide questions you already settled, does not overstate what your project actually does, and never edits its own core vision without your explicit approval. You install it by copying the starter pack folder into your Claude Code skills directory, then running a single command inside any project, either new or existing. Claude Code then asks you questions about your project, including your vision for it in your own words, your tech stack, and your branching approach. Nothing gets written into the vision document until you approve it line by line. If you run it on a project that already has some of these files, it only fills in what is missing rather than overwriting anything. The scaffolded files each cover a different kind of information: a vision file for your core intent, a rules file with standing behavioral guidelines, an architecture file for your design decisions, a strategy ledger for decisions and rejected options that gets read into every new session automatically, a short status snapshot, a dated history log, a roadmap with checkboxes, a lessons learned file, and append-only logs of decisions and errors. Beyond just writing these files, the project adds real enforcement. The vision file is made read only on disk, and a pre-commit hook blocks any attempt to commit changes to protected files, though you can still override this deliberately if needed. This is meant to stop an agent from simply forgetting or ignoring its own guardrails over time. The project is released under the MIT license.