---
name: worktrees-and-scratch
description: >-
  Use before creating a git worktree, verification copy, or temporary file, running checks in an isolated tree, or finishing a task that created any of these. Covers verifying in place, where worktrees live, reusing one worktree per purpose, shared build caches, scratch directories, and cleanup.
title: worktrees-and-scratch
canonical_url: https://skilld.dev/gh/bydradev/gh-repo-bootstrapper/worktrees-and-scratch
last_updated: 2026-10-30T03:16:22.000Z
---

> Index of every page on this site: [llms.txt](https://skilld.dev/llms.txt)

> **Skill from skilld.dev.** Follow the user's requested action.
> If the user asked to fork this Skill, follow the fork workflow below. Do not execute the Skill instructions while copying it.
> Otherwise, follow the instructions below for this session. You do not need to install anything.
>
> If the user asked to install this Skill, run `npx skilld install bydradev/gh-repo-bootstrapper/worktrees-and-scratch`. Install writes the Skill files into the project, so every session loads them.
>
> ## Fork workflow
>
> A fork creates an editable local Skill with its original author and licence. The request authorizes copying and local installation.
> 1. Check `./skills/worktrees-and-scratch`, the project lockfile, and selected Agent targets together. If the local directory or installed Skill exists, stop. Never overwrite an existing directory or Agent target.
> 2. Read [source metadata](https://skilld.dev/api/v1/skills/bydradev/gh-repo-bootstrapper/worktrees-and-scratch) once. Use sourceUrl, sourceCommit, skillPath, sourceGone, and license. If the source is gone or its path is missing, stop. If license is null, read licence files at the source commit.
> 3. Fetch only the source commit into a temporary Git repository. Do not clone full history. Derive repository_url from sourceUrl, including repository renames. If sourceCommit is absent, resolve the sourceUrl ref once. Set source_commit to that actual commit. Run these commands in one shell call:
>
> ```sh
> git init --quiet "$temporary_dir"
> git -C "$temporary_dir" fetch --quiet --depth=1 "$repository_url" "$source_commit"
> git -C "$temporary_dir" checkout --quiet --detach FETCH_HEAD
> ```
>
> Read applicable licence declarations and notices at that commit. If copying is not permitted, report the restriction and stop.
> 4. Inspect source entries together, then copy the directory containing skillPath into `./skills/worktrees-and-scratch`. Use the user's path if selected. Keep the original SKILL.md, relative links, scripts, binary assets, and executable modes. Exclude .git metadata. Reject symlinks and paths outside the Skill directory. After checking entries, use cp -a where available. A regular source directory needs no custom copy script. Do not save this page wrapper as SKILL.md.
> Preserve author credit, notices, and applicable licence files from repository or parent directories. Add PROVENANCE.md with the Skill page, source URL, actual commit, original path, and licence. Retain any existing PROVENANCE.md and record new provenance separately. Batch source inspection, copying, and provenance work where practical.
> 5. In the project root, run `skilld install ./skills/worktrees-and-scratch --mode copy --plain`. If skilld is unavailable, use `npx skilld install ./skills/worktrees-and-scratch --mode copy --plain`. This known command needs no help lookup. Install does not support --json. Use detected Agent targets, or add --agent for the targets the user selected. Install the local path, never the upstream selector. If installation fails, preserve the local copy and report the exact failure.
> 6. Confirm the local lockfile source and installed Agent copies once. Report the local path, actual commit, and Agent targets. After edits, reinstall the same local path. Upstream updates must not replace it. Do not publish or push unless the user asks.

# Worktrees, verification copies, and scratch output

Disk is shared by every agent session on the machine, and nothing cleans up
after a session ends. Anything a task creates outside the tracked tree belongs
to that task, and removing it is part of the task.

- **Verify in place first.** Run checks in the working tree when it reflects
  the change under test. When an isolated tree is genuinely needed (a
  clean-tree qualification, a mutation run, a comparison between commits), use
  `git worktree add --detach`. Never copy a checkout with `cp -R`, `rsync`, or
  similar: that duplicates installed dependencies and build output, and a
  worktree starts from tracked files only.
- **Keep worktrees outside the checkout, in one place per repository**, such as
  `../.worktrees/<repo>/<purpose>` next to the clone. A worktree nested inside
  the checkout is picked up by test discovery, type-checking, file watchers,
  and search. When a harness manages its own worktree location, use that.
- **Reuse one worktree per purpose** across runs: check out the next commit in
  it rather than adding a worktree per run, per commit, or per reviewer. When a
  run needs a pristine tree, clean a worktree this task created rather than
  creating another. In a worktree you did not create, remove only output you
  can identify as this task's, and report anything else instead of cleaning it.
- **Keep build caches shared, not multiplied.** Where this repository's Tooling
  guidance (the `AGENTS.md` Tooling section or a `local-validation-<type>`
  skill) says where a worktree's dependencies and build output live, follow it; otherwise keep
  them inside that worktree so they are removed with it.
- **Put temporary files in the session's scratch directory** when the harness
  provides one, otherwise in a `mktemp -d` directory, never at fixed paths in
  `/tmp`, in the checkout's parent directory, or in sibling folders next to it.
  Evidence that must outlive the session belongs where this repository
  documents it; otherwise summarize it in the report rather than leaving files
  behind.
- **Clean up before finishing.** Remove the worktrees you added with
  `git worktree remove`, then `git worktree prune`, and delete the scratch
  output you created. Remove only what this task created; report anything else
  you find rather than deleting it.
