# Merge

> The user-triggered finisher that executes the post-PR tail per `.yoke/flow.md`: merges the task's PR(s), runs cascade merges, runs deploy/release commands, moves the ticket to its target state, cleans up worktrees, and returns to the default branch. Activates when the user writes "merge", "merge it", "/merge", "finish the task", "ship it", "cascade and deploy", or after the user approves a PR on GitHub. Never runs on its own — the merge decision stays the user's.

- Skill: `yokeloop/merge` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add yokeloop/merge`
- Raw SKILL.md: https://api.skillmd.com/api/skills/yokeloop/merge/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: yokeloop (https://skillmd.com/u/yokeloop)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/yokeloop/merge

---


# Finish the task: merge, cascade, deploy, transition, clean up

The flow ends at PR (ADR-0006): `do` drives every run to a ready pull request and
stops there. The merge decision is the user's, made on GitHub. `/merge` turns that
one decision into one command — it executes the whole post-PR tail per the flow
map instead of the user dictating the same paragraph every session.

**Principle:** merging is the user's call; `/merge` runs only on the explicit word.

---

## Input

`$ARGUMENTS` — optional PR URL(s) or a `<slug>`. Empty means the current
branch/worktree's task: resolve its open PR(s) via `gh` (e.g. `gh pr view`,
`gh pr list --head <branch>`).

---

## Procedure

Read the flow map once at the project root:

```bash
bash ${CLAUDE_PLUGIN_ROOT}/lib/flow-read.sh
```

Then read and follow `reference/merge-procedure.md` — the one reference this skill
routes to for the full mechanics. It runs this step sequence, each step **skipped
silently when the flow map does not declare it**:

1. **Merge PR(s)** — merge the task's pull request(s) on GitHub.
2. **Cascade** — advance the branch cascade (e.g. master → staging → develop) by
   each declared step command.
3. **Deploy / release** — run the `DEPLOY_COMMANDS` the flow map declares, nothing
   inferred.
4. **Ticket transition** — move the task's ticket to its target state.
5. **Worktree cleanup** — remove the task's worktree(s) and delete merged branches.
6. **Return to the default branch** — sync and land on the repo's default branch.

---

## Rules

- **Never auto-runs.** `/merge` fires only on an explicit user command — no
  automatic trigger after `do`, ever. The merge decision stays the user's.
- **Skip undeclared steps silently.** A step whose section is absent from the flow
  map is skipped without nagging — omission means "apply the default", never "fail".
- **Multi-repo.** Merge every one of the task's PRs; libraries were already
  direct-pushed by `do`, so there is nothing to merge for them.
- **Git authorization** comes from the skill invocation itself — see "Git initiative
  and defaults" in `${CLAUDE_PLUGIN_ROOT}/skills/gca/reference/commit-convention.md`.
  Do not restate it; do not ask "merge?" mid-run.
- **Artifacts only under `.yoke/`.**
- **Any fork** — an ambiguous PR set, a merge conflict, an unclear target — is one
  AskUserQuestion with a recommended option listed first.
- **Language:** match the ticket/input language, or follow the project-level
  definition in CLAUDE.md / AGENTS.md.

---

## Notify

On finish, send the terminal STAGE_COMPLETE notification with what was merged
and deployed — the exact call lives in `reference/merge-procedure.md` §7.