laurigates/configure-worktreeinclude

/configure:worktreeinclude

Generate a .worktreeinclude file whose patterns tell Claude Code which gitignored files to copy from the main checkout into a newly-created git worktree. A worktree is a fresh checkout containing only git-tracked files, so gitignored runtime inputs — .env, local secrets, local config — do not come along by default; this skill builds the include list from the gitignored files actually present in this repo, so the patterns match the project instead of a generic template.

When to Use This Skill

| Use this skill when... | Use another approach when... | |------------------------|------------------------------| | Worktrees / subagent worktrees start without .env or local config | Ignoring a build artifact — use /configure:gitignore | | Onboarding a repo to Claude Code (also reachable via /configure:repo) | Copying large regenerable dirs (node_modules/) — let the worktree's setup hook reinstall them | | You want gitignored inputs reproduced per-worktree without committing them | A non-git VCS — a custom WorktreeCreate hook must copy files itself; .worktreeinclude is git-only |

What .worktreeinclude is

.worktreeinclude lives at the repo root, alongside .gitignore, and uses .gitignore glob syntax. When Claude Code creates a worktree (the --worktree flag, subagent worktree isolation, or parallel desktop sessions), every file that matches a pattern and is gitignored is copied into the new worktree. The match is constrained two ways by design:

• Only gitignored files are copied. A pattern that matches a tracked file is a no-op — tracked files are already in the worktree, never duplicated.
• You choose the inputs, not the bulk. The right entries are small per-environment inputs (env files, local secrets, local config). Large regenerable trees (node_modules/, .venv/, build output) are better rebuilt by the worktree's setup than copied — copying is slow and can carry platform-specific binaries or absolute paths.

Commit .worktreeinclude so the whole team (and every agent worktree) shares the same include list.

Context

• Existing worktreeinclude: !find . -maxdepth 1 -name .worktreeinclude -type f
• Gitignore present: !find . -maxdepth 1 -name .gitignore -type f
• Ignored entries present (collapsed): !git status --ignored --porcelain=v1

Parameters

| Parameter | Description | |-----------|-------------| | --check-only | Report the proposed include patterns without writing any file | | --fix | Write the proposed patterns without prompting (headless) |

Default (no flag): propose the patterns, confirm the set with the user via AskUserQuestion, write .worktreeinclude, and git add it (staged, not committed).

Execution

Execute this .worktreeinclude configuration workflow:

Step 1: Read existing state

Parse $ARGUMENTS for --check-only / --fix. From Context, note whether .worktreeinclude already exists (Read it and record its current patterns if so) and capture the ignored-entry list. If no .gitignore exists and no ignored entries are present, report "nothing gitignored to include" and stop.

Step 2: Classify the gitignored entries

The ignored list comes from git status --ignored --porcelain=v1 — lines prefixed !! are ignored paths, with directories collapsed to a single trailing / entry. Sort each entry into one of two buckets:

| Propose to include (small per-environment inputs) | Skip (regenerable / large / risky) | |--------------------------------------------------------|-----------------------------------------| | .env, .env.local, .env.* (not a tracked .env.example) | node_modules/, .venv/, venv/, .direnv/ | | *.local config (config.local.*, settings.local.*) | dist/, build/, out/, target/, .next/, .nuxt/ | | local secrets / credentials present as files (secrets.*, gitignored *.pem/*.key) | __pycache__/, .pytest_cache/, .ruff_cache/, .mypy_cache/ | | gitignored tool config the build needs (.npmrc, .tool-versions) when present | coverage/, .coverage, *.log, .DS_Store, .git/ |

Only propose patterns for entries that actually appear in the ignored list — the include list must describe this repo. When in doubt about a file (it could be a needed input or just local cruft), surface it as a candidate rather than deciding silently.

Step 3: Confirm the set

• --check-only: print the proposed patterns as a diff-style block; write nothing.
• --fix: skip confirmation; use the proposed include set as-is.
• Default: present the proposed patterns with AskUserQuestion, letting the user drop any pattern or add one you skipped (the built-in "Other" option covers additions). Keep the question self-contained — list each proposed pattern and the file(s) it matches.

Step 4: Write or report

Write the confirmed patterns to .worktreeinclude at the repo root, one pattern per line, with a short header comment. If the file already exists, merge additively — preserve existing lines verbatim and append only patterns not already present (idempotent, safe to re-run). Then git add .worktreeinclude.

Step 5: Verify

Confirm each written pattern targets a gitignored file (so it will actually be copied): git check-ignore -v <path> should report a match for a representative file behind each pattern. A pattern that matches no gitignored file is inert — flag it so the user knows it does nothing.

Step 6: Report

Print a summary: patterns added vs already-present, the file path, and the git diff --cached hint. If --check-only, prefix with "DRY RUN — no files modified".

Important Notes

• Built from this repo, not a template. Every proposed pattern must trace to a gitignored file present in the working tree — that is what "based on the repo in question" means.
• Additive only. Never remove or reorder existing .worktreeinclude lines; append missing patterns so the skill is idempotent.
• Patterns only match gitignored files. A pattern over a tracked file is a silent no-op — git check-ignore in Step 5 catches inert patterns.
• Prefer rebuild over copy for big trees. Keep node_modules/, virtualenvs, and build output out of the include list; let the worktree's setup hook regenerate them.
• Commit the file. .worktreeinclude is shared configuration — stage it and let the user commit so every worktree (and teammate) uses the same list.
• Never auto-commit; stage and let the user review with git diff --cached.

Agentic Optimizations

| Context | Command | |---------|---------| | Propose + write the include list | /configure:worktreeinclude | | Dry-run, no file changes | /configure:worktreeinclude --check-only | | Write headless (no prompt) | /configure:worktreeinclude --fix | | List gitignored entries (collapsed) | git status --ignored --porcelain=v1 | | Confirm a pattern matches a gitignored file | git check-ignore -v .env | | Review staged change | git diff --cached |

See Also

• /configure:gitignore — the sibling skill that keeps Claude Code runtime state (incl. .claude/worktrees/) out of commits
• /configure:repo — full repo onboarding (offers this skill)
• Claude Code worktrees documentation