adhoc-pr

name	adhoc-pr
description	Ad-hoc Pull Request Publisher — commits the current changes, pushes the branch, opens a GitHub pull request against the repository's default branch, and creates a matching SSP Jira issue in the current sprint following the ad-hoc work convention (summary = PR title + " #<PR number>", PR link in the "Merge Request" field, no description, moved to In Progress). Works for any ad-hoc change — fixes, small improvements, tooling, docs. Use when the user asks to publish/ship the current work end-to-end, or invokes /adhoc-pr.
user_invocable	true
version	1.0.0

Ad-hoc Pull Request Publisher

Wraps the whole "ad-hoc change" flow: commit → push → GitHub PR → Jira issue in the current sprint. Ad-hoc work (a fix, a small improvement, tooling, docs, …) doesn't get a spec — the PR is the documentation; the Jira issue only makes the work visible in the sprint.

Assumes an authenticated GitHub CLI (gh) and a connected, authorized Atlassian (Jira) MCP server — no upfront verification is done; if a step fails on missing tooling, report what was completed and the user finishes the rest manually.

Command Arguments

• [issue type] (optional): defaults to PRG bug for fixes; use PRG US for improvements/features and FE bug/FE US for storefront-only work
• check (optional flag): also run the full make check-fix before committing

The Jira project is always SSP — it is not configurable.

Workflow

Step 1: Preconditions

git status --short                                            # what will be committed
git branch --show-current                                     # current branch
gh repo view --json defaultBranchRef --jq .defaultBranchRef.name   # default branch (do NOT hardcode)
git fetch origin <default-branch>                             # make the comparison ref current

• If the working tree is clean and the branch is already pushed with an open PR, skip the finished steps and continue with the first missing one (idempotent re-run).
• If the current branch is the default branch, create a work branch first: <initials>/<kebab-slug> (initials from git config user.name, e.g. rv/…).
• The branch must be based on the default branch — verify with git log --oneline origin/<default-branch>..HEAD (against the just-fetched remote ref, never a possibly stale local one) that only the intended commits would enter the PR. If foreign commits appear, rebase onto origin/<default-branch> and remember that a rebase happened — Step 3 then needs --force-with-lease.

Step 2: Commit

1. make check-fix is not run by default — it takes minutes even for trivial changes, and the user is expected to have handled code quality beforehand. Run the full make check-fix only when the user requests it (the check argument or an explicit ask). If the changes look risky for standards (larger PHP/TS code changes, generated files possibly stale), offer it via a quick question rather than running it unasked. When it does run: fixes it makes belong in the commit; on failure stop and report.
2. Invoke the commit skill (Skill tool) to create the commit(s) from uncommitted changes. Skip if everything is already committed.

Step 3: Push

git push -u origin <branch>

If Step 1 rebased an already-pushed branch, push with --force-with-lease instead — never plain --force, and never force-push a branch someone else may have built on without confirming first.

Step 4: Pull Request

1. Check for an existing PR before generating anything: gh pr view --json number,title,url,state. Reuse it only when state is OPEN — a MERGED/CLOSED PR belongs to previous work; if the branch's PR was already merged, stop and report instead of attaching new work to it.
2. If no open PR exists, invoke the pr-description skill to craft the title and description (business-value focused, uses .github/PULL_REQUEST_TEMPLATE.md).
3. Write the body to a file with the Write tool (e.g. into the scratchpad directory) and create the PR with --body-file — never inline the body in double quotes, PR bodies in this repo routinely contain backticks that the shell would interpret as command substitution:

gh pr create --base <default-branch> --head <branch> --title "<title>" --body-file <path>

Record the PR title, number, and URL — the Jira issue is derived from them.

Step 5: Jira Issue

Use the Atlassian MCP tools with cloudId: shopsys.atlassian.net.

Field reference (SSP project, board 167):

Field	Id	Value
Sprint	customfield_10020	numeric id of the active sprint
Merge Request	customfield_10031	the GitHub PR URL

1. Active sprint id: search JQL project = SSP AND sprint in openSprints() ORDER BY updated DESC with maxResults: 1 and fields: ["customfield_10020"]; from the first result take the customfield_10020 entry whose state == "active" (an issue can carry closed sprints too; for SSP the active one has boardId 167). If the search returns no issues (fresh sprint with nothing in it yet), create the issue without the sprint field and tell the user to drag it into the sprint manually — do not guess a sprint id.
2. Assignee: the requesting user — get account_id via atlassianUserInfo (or reuse a known one from context).
3. Create the issue (createJiraIssue):
   • projectKey: SSP
   • issueTypeName: PRG bug (or the argument)
   • summary: <PR title> #<PR number> — exactly the PR title, then space and #<number> (e.g. Fix expanding lazily loaded branches in admin tree selection #4701)
   • no description — the convention is to leave it empty
   • additional_fields: {"customfield_10020": <sprintId>, "customfield_10031": "<PR URL>"}
   • assignee_account_id: the user's account id
4. Fallback: if creation rejects customfield_10031 (field not on the create screen), create the issue without it and set it afterwards via editJiraIssue with fields: {"customfield_10031": "<PR URL>"}.
5. Move to In Progress: the work is already done, so the issue must not sit in the backlog column. Call getTransitionsForJiraIssue on the new issue, pick the transition whose target status (to.name) is In Progress (case-insensitive match on the status name, not the transition name), and apply it with transitionJiraIssue. If no such transition is offered, leave the issue in its initial status and report that to the user instead of guessing another status.
6. Verify: fetch the new issue directly with getJiraIssue (not a JQL search — search indexing can lag behind for just-created issues) requesting fields: ["summary", "status", "customfield_10020", "customfield_10031", "assignee"], and check the sprint, the Merge Request link, the summary suffix, and that the status is In Progress.

Step 6: Report

Present to the user: commit hash(es), PR URL, Jira issue key + URL (https://shopsys.atlassian.net/browse/<KEY>), and the sprint it landed in.

Rules

• Never hardcode the default branch — resolve it each run; it changes with major releases (19.0 → 20.0 → …).
• The Jira summary must copy the PR title verbatim and end with #<PR number>.
• The PR link belongs in the Merge Request field, not in the issue description.
• Transition the issue to In Progress (assigned to the author) — never further; CR, testing, and BV statuses are driven by the team workflow, not by this skill.
• If any step fails, stop and report what was already done (commits, PR, issue) so the user can pick up manually — do not retry destructively.