Spin ONE idea off into its own parallel Claude session as a throwaway branch, while THIS session keeps working — and DON'T merge it back on finish. For a spur-of-the-moment "what if" you want out of the way of the main work and may bin after a look: the orchestrator commits a clean START, hands you one paste-and-forget block for a new session, and polls a shared status file so you relay nothing. When the offshoot reports done it HOLDS — surfaces its branch and any screenshot/preview for you t...
Scanned 9/2/2026
Install to Claude Code
npx -y skills add skowalskidev/dotclaude --skill work-split-session-in-parallel-branch-offshoot --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Work Split Session In Parallel Branch Offshoot?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/skowalskidev-work-split-session-in-parallel-branch-offshoot)More formats (shields.io, HTML) on the badges page.
---
name: work-split-session-in-parallel-branch-offshoot
description: Spin ONE idea off into its own parallel Claude session as a throwaway branch, while THIS session keeps working — and DON'T merge it back on finish. For a spur-of-the-moment "what if" you want out of the way of the main work and may bin after a look: the orchestrator commits a clean START, hands you one paste-and-forget block for a new session, and polls a shared status file so you relay nothing. When the offshoot reports done it HOLDS — surfaces its branch and any screenshot/preview for you to eyeball — and waits for your call: merge it back, or discard it (branch + worktree binned). Reuses /sk:work-hyperspeed's START/paste/poll plumbing; the only difference is the finish is hold-don't-assemble, and it is always ONE session, not a 3-5+ fan-out. Use for "work this as a branch offshoot", "spin this off, I might discard it", "try this as an offshoot in parallel", "explore this on a side branch I'll eyeball later", "offshoot this", "don't merge it back yet". For a task that DIVIDES into 3-5+ slices, that is /sk:work-hyperspeed instead.
argument-hint: "[the idea to spin off as a throwaway branch]"
---
# Branch offshoot — one parallel spike you hold, then merge or bin
An offshoot is a single idea run in its OWN session, off to the side of the main work, that you look at
then keep or throw away. It is NOT hyperspeed: hyperspeed DIVIDES a task across 3-5+ sessions and
ASSEMBLES them the moment they finish. An offshoot is ONE session and DOES NOT auto-merge — the point
is that you decide, after eyeballing it, whether it lives.
## When to use it — and when not
DO reach for it when Simon says "spin this off / try it as an offshoot / branch off and I'll look later
/ I might discard it" — a self-contained "what if" he wants OUT of the main session's way and is willing
to bin.
DON'T use it for a task that divides into 3-5+ independent slices — that is `/sk:work-hyperspeed`.
DON'T use it for work that must land — an offshoot's default fate is the bin, so a change Simon needs
merged goes through `/sk:work-full-detailed-workflow` on this branch, not here.
## Reuse the plumbing — only the finish differs
The START-commit, the single paste-and-forget block, and the shared-status POLL are the shared handoff
unit — `references/parallelization.md` § "The hand-run session handoff", run with exactly ONE part and
`<harness>` = `offshoot`; do not restate them. Two deltas for an offshoot:
- **The part is ONE, and it does not assemble.** Cut no partition — the whole idea is the single part.
Its paste block still branches off START, writes `working`→`done` to the run's STATUS_DIR, and reports
its branch plus the absolute path of anything viewable it produced (a screenshot set, a preview
build), the way hyperspeed's standalone-artifact variant does.
- **Build it to a LOOKABLE state, not a merged one.** The part builds/runs far enough that Simon can
eyeball the idea (a before/after, a screenshot, a running route), then STOPS. It opens no PR and the
orchestrator does not merge it.
## Finish — HOLD, then merge or bin on Simon's word
When the status file reads `done`, do NOT assemble. Instead, per `references/human-pacing.md`:
1. **Surface it for the eyeball.** Fetch the branch, gather the reported artifact into this session's
scratch, and SEND it to Simon (screenshots via SendUserFile, or the diff summary). One line: what it
is, its branch name, that nothing is merged.
2. **WAIT for his verdict — never merge on your own.**
TEST: after `done`, the offshoot branch is neither merged nor deleted until Simon gives an explicit
merge-or-discard call.
3. **On "merge it":** `git fetch`, merge the offshoot branch into the working branch, run the project
gate (build + tests), then offer cleanup as in hyperspeed's Step 6 (delete the branch local+remote
after proving it merged; the part's worktree goes when Simon archives that session).
4. **On "bin it":** delete the offshoot branch (local + remote) and its worktree WITHOUT merging —
confirm the exact list first (`references/git-pr-deploy.md` § deleting a branch), then remove. A
discarded offshoot leaves no branch, no worktree, no status dir behind.
Hold indefinitely if he says neither — an offshoot has no deadline; it waits until he returns to it.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!