agentsclimarketplace

Harness worktree

Skill cekrauseee/harness/skills/harness-worktree

File-native project continuity and engineering standards for coding agents.

Install
npx -y skills add cekrauseee/harness --skill harness-worktree

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 23 days oldThe repository was created 23 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Plan, create, validate, adopt, and retire isolated Git worktrees with Harness-controlled semantics and lifecycle while delegating only physical path and storage to the active agent host. Use when naming, creating, locating, validating, adopting, reusing, or removing a worktree, or when a task needs the Harness `type/slug` branch convention.

SKILL.md

4.5 KB, 837 tokens by cl100k_base, as published. Nobody here has run it

Harness Worktree

Apply a Harness-controlled creation protocol and lifecycle while using the active host's storage convention. Harness owns branch naming, base selection, creation, isolation, validation, adoption, reuse, and retirement. The host owns only the physical checkout path, storage layout, and native bookkeeping required to integrate that path.

Plan a Worktree

  1. Read applicable repository instructions and identify the task's primary intent.

  2. Choose an allowed Conventional Commit type and translate the task into a short English kebab-case slug. Read references/worktrees.md when classification or collision behavior is unclear.

  3. Run:

    python3 scripts/resolve_branch.py \
      --project <checkout> \
      --type <type> \
      --slug "<short English description>" \
      --require-available
    
  4. Use the returned branch and validated base exactly. The resolver derives the remote default or current branch when --base is omitted. It never chooses a worktree path.

  5. If the branch exists, adopt it only when the user requested adoption and its Git identity matches. Otherwise choose a distinct slug; never silently rename or reuse an existing branch.

Create with Host Storage

  1. Detect how the active host allocates and records worktree paths from system instructions, available tools, and repository guidance.
  2. Obtain or accept the path selected by that host. In Codex, use Codex path allocation; in Claude Code, use Claude Code path allocation. Never substitute a Harness global path.
  3. Create the worktree from the resolved base on the resolved branch. Prefer the host's native operation when it preserves that plan and native bookkeeping; otherwise use ordinary Git at the host-selected path.
  4. When a native operation combines path allocation and creation, treat it as the execution mechanism for the Harness plan, not as ownership of the creation protocol or lifecycle.
  5. Validate the result before starting task work. If the host cannot provide compatible storage or preserve the Harness plan, explain the incompatibility instead of weakening the semantics.

Harness provides no fallback storage root. If the host exposes no allocator, follow explicit user, repository, and host path conventions while keeping the path outside Harness global state.

Validate Isolation

After creation or adoption:

  • confirm the checkout is registered by git worktree list --porcelain;
  • confirm its current branch is the resolved type/slug branch and its base is the planned revision;
  • keep implementation, generated files, dependency setup, and verification inside the worktree;
  • run the relevant setup and tests from the worktree, then report the host-selected path and results;
  • leave the source checkout and unrelated worktrees unchanged.

Control the Lifecycle

  • Adopt or reuse an existing worktree only when requested and after its repository, branch, base, and path identity match the plan.
  • Keep the host-selected path in session or handoff context when continuity requires it; do not turn that path into a Harness storage convention.
  • Before retirement, inspect worktree registration and status. Never remove a checkout with uncommitted or untracked work.
  • Remove or prune only with explicit authorization. Prefer the host's native removal operation when it updates host bookkeeping; otherwise use ordinary Git.
  • Confirm retirement removed only the intended registration and checkout. Leave the branch intact unless branch deletion was separately authorized.

Mutation Boundary

  • Resolving branch semantics does not authorize worktree or branch creation, removal, pruning, or deletion.
  • Materialize the plan only when the user explicitly asks to create the worktree.
  • Never remove or overwrite a worktree with uncommitted changes.
  • Do not create Harness metadata inside the repository.

Portable Semantics

branch:   <type>/<short-kebab-case-slug>
base:     <validated repository revision>
checkout: isolated, host-stored, Harness-managed

Host, agent, and user names never appear in the Harness branch. Worktree directory names and locations are selected by the host and are not part of the portable Harness contract.

What ships with it: 3 files

6.6 KB alongside SKILL.md, 1 of them executable

agents/

references/

scripts/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.