Auto-Detect

The AI stages your modified files, generates a descriptive message that includes what changed and why, commits, runs a post-commit review, and optionally pushes to GitHub with automatic token scrub. Other recipes call this one automatically — you rarely need to invoke it directly.

The AI checks whether git is already configured, initializes a repository if needed, sets your identity, creates a CRAFT-aware .gitignore, makes an initial commit, and optionally connects a GitHub remote — all idempotent, so running it twice won’t break anything.

The AI captures a snapshot of your session progress so far — what you have accomplished, which files were touched, and your current priorities. It writes this as a partial handoff to your chat history file (or updates an existing partial if one already exists from an earlier checkpoint). A lightweight validation confirms the checkpoint data matches actual git state. Then it creates a git commit so the checkpoint is recoverable even if the session ends unexpectedly.

The AI scans your workspace folder, finds your project files, loads your last handoff, checks git status, activates your persona, and reports everything it found — so you can start working immediately without manually pasting context.

The AI compiles your session data (accomplishments, files touched, key decisions, next priorities), generates a structured handoff block in CRAFT format, and runs an integrated Lessons Learned capture that identifies new learnings, formats them, presents them for your review, and writes approved entries to your LL file. Before committing, the handoff passes through a validation gate where claims are independently verified against git history and file state using Chain-of-Verification. The validated handoff is appended to your chat history file and a git checkpoint is created.