git

Use for git history and code navigation -- scoped commits with git codex, pushes, PRs, merges, read-only worktree inspection, or impact analysis. The active project instructions govern mutation authority.

methodology

1. read git status and the live project instructions; stage only owned paths. commit with git codex -m "type(scope): description", leaving the normal git user as committer; add no agent attribution to commits or PR text.
2. use git for authorized history work. its broad staging, plain commit, rebase, checkout, and merge examples yield to the project rules and the trained decisions.
3. use worktree for read-only inspection; create, remove, or prune only with explicit authority.
4. for code questions, use the repo's codegraph first. gkg is an optional historical source when explicitly requested and available.
5. guide instructions are adapted to the live tools and authority they run under; the router and trained layer hold cross-guide decisions.

contents

• trainedlearned layer -- read first
• gitcommits, pushes, PR creation and merge, branch merges, PR review-and-merge automation, CI follow-up, secret scanning
• gkgsemantic code analysis with GitLab Knowledge Graph -- go-to-definition, find-usages, impact analysis, architecture visualization (Ruby, Java, Kotlin, Python, TS/JS)
• worktreecreating, inspecting, and cleaning isolated git worktrees -- feature isolation, health audits, stale cleanup, monorepo/submodule workflows

procedures

procedures the guru runs, one page each. read it, then follow it.

• ck-worktreecreate isolated git worktree for parallel development
• git-cmstage all files and create a commit
• git-cpstage, commit and push all code in the current branch
• git-prcreate a pull request
• git-prepare-repoclean up and prepare a git repo for client delivery -- remove unwanted tracked files, trim repo size, improve documentation