Project Initialization

Project Initialization

Initialize the project in the current working directory. Existing project files are preserved in an old* directory for later review; do not merge them into the new scaffold during this workflow.

Dual-Runtime Project Generation

The bundled template contains Claude project sources under .claude/**. Its pre-commit hook runs the installed ~/.claude/scripts/sync-to-codex.sh converter after those sources are staged. This generates and stages the matching AGENTS.md and .codex/** runtime before the commit while .gitignore excludes host-local .codex/.sync/**. A reported conflict or validation error stops the commit.

1. Check the Current Directory

Work only in the current directory. If it is already a Git repository with uncommitted changes, ask whether to commit them first, continue while preserving them in old*, or stop. Do not proceed until the user chooses.

Resolve the directory of this loaded project-initialization skill and set INIT_SKILL_DIR to that absolute path. Verify that its assets/new-project/ directory exists before moving project files.

2. Preserve Existing Files and Apply the Template

If the current directory contains anything other than .git, move all such entries into the first available directory named old, old2, old3, and so on. Do not move .git.

OLD_DIR=""
if find . -mindepth 1 -maxdepth 1 ! -name '.git' -print -quit | grep -q .; then
  OLD_DIR="old"
  N=2
  while [ -e "$OLD_DIR" ]; do OLD_DIR="old${N}"; ((N++)); done
  mkdir "$OLD_DIR"
  find . -mindepth 1 -maxdepth 1 ! -name '.git' ! -name "$OLD_DIR" -exec mv -- {} "$OLD_DIR/" \;
fi

cp -rp "$INIT_SKILL_DIR/assets/new-project/." .

If OLD_DIR is non-empty, inspect it for .env*, *.key, *.pem, credentials.json, and secrets/. Ensure every sensitive path is covered by the new .gitignore before staging files. Never print secret contents.

3. Initialize Git and Hooks

Initialize Git with main as the primary branch when needed. For an existing repository, preserve its history and make the current primary branch main. Then:

1. Make .githooks/pre-commit executable.
2. Set git config core.hooksPath .githooks.
3. Stage the scaffold and preserved old* directory, subject to the secret check above.
4. Create the first initialization commit. The pre-commit hook generates and stages AGENTS.md and .codex/** while .gitignore excludes .codex/.sync/**. If generation reports a conflict or validation error, the commit stops. If the repository already has history, create a normal initialization commit instead of rewriting existing commits.

4. Connect GitHub and Create Branches

GitHub is required for this workflow. Verify that gh is installed and authenticated.

• If origin already exists, show its URL and ask whether it is the intended repository. Stop on a mismatch rather than replacing the remote. Confirm through gh that the repository is private; if it is public, stop and ask whether to make it private or use another repository.
• If origin does not exist, ask for the GitHub repository name unless already supplied. The supplied name authorizes creating the private repository. Create it with gh repo create {name} --private --source=. --remote=origin.

Before any push, inspect whether a local or remote dev branch already exists. If neither exists, create dev from main. If either exists, show its relationship to main and ask whether to reuse it; never reset or recreate an existing dev, and stop if local and remote histories conflict.

Ask explicitly before pushing main. After approval, push main, push the created or approved dev, leave dev checked out, and read the canonical repository URL through gh.

5. Report the Result

Report:

• the GitHub URL;
• the created or reused main and dev branches;
• the preserved old* directory, or that the directory was initially empty;
• that dev is the active branch;
• the next step: create the initial Project Knowledge with documentation-writing.

Do not review or merge files from old* during initialization.