Resolve a single Liferay test failure end-to-end.
Create a Jira bug ticket in the LPD project through the REST API. Use when the user asks to create or file a Jira bug or LPD ticket.
Create a Jira task in the LPD project through the REST API. Use when the user asks to create a Jira task or LPD task ticket.
Format source files to align with Liferay's coding standards.
Create a GitHub PR for the current branch. Use when the user asks to create a PR, send a PR, or invokes /pr.
Start work on a Jira ticket.
| allowed-tools | ["Agent","Bash","Edit","Glob","Grep","Read","Skill","Write"] |
| description | Check that a PR is ready to be sent for review. |
| name | pr-check |
Run premerge checks against the current branch. The skill iterates through the validations under validations, runs each one whose trigger matches the diff, and reports PASS or FAIL. Integration tests, Playwright tests, and Poshi tests are out of scope. Use the test-plan skill when their coverage is needed.
Branch rebased on local master. Compare git merge-base HEAD master to git rev-parse master; when they differ, abort and tell the developer to rebase (git rebase master). The format-source-current-branch validation diffs against the local master tip to decide which files to format — when the branch's merge-base lags behind that tip, SF picks up reverse-direction diffs from the master commits the branch has not absorbed and the <TICKET> SF autocommit captures unrelated rewrites.
Working tree clean. git status --porcelain must return empty. When dirty, abort and ask the developer to commit first — the autocommit steps inside drift validations would otherwise capture their uncommitted edits.
Diff baseline is local master. The skill never fetches from a remote and does not compare against an upstream/master ref. Whether local master is up to date is the developer's call.
Diff is nonempty. When the three-dot diff produces no files, exit with a one-line message — no validation produces useful signal on a clean branch.
git diff --name-status "$(git merge-base HEAD master)...HEAD"
PASS or FAIL.
The procedure runs in two passes over the validations, in the order below. The order is dependency-driven: drift first (later validations see the regenerated tree), then formatting, then build, then tests.
Process each validation in a subagent.
Read every validation file under .claude/skills/pr-check/validations in a single parallel batch — one Read tool call per file, all in the same tool-use turn. From each file, take the regex inside its ## Match section.
In your next turn, compose a single bash script that:
git diff --name-only "$(git merge-base HEAD master)...HEAD"! in the regex inverts: fire when any diff path does not match the rest)From the script's output, sum the matched validations' ## Time Estimate values for the cumulative total. The matching is mechanical; consult each file's prose ## Trigger only when a result needs human-judgment context (e.g., Service Builder output-only catch-up).
When the total exceeds 20 minutes, surface the breakdown and ask the developer whether to trim a validation or proceed.
For each matched validation, spawn one subagent. Pass it only the ## Command and ## Autocommit sections of the validation file, not the full file. Record PASS or FAIL. Do not halt on FAIL — continue so the developer sees the full picture.
When the validation's Command is a build (gradle, ant, npm, jest), bound the output:
<command> 2>&1 | tail --lines=100
Decide PASS/FAIL from the build tool's success markers in the captured output (BUILD SUCCESSFUL / BUILD FAILED, Tests: N passed, M failed, etc.). Apply only to build commands. Leave inert commands like git status --porcelain and git diff --quiet untouched.
When all validations pass, report PASS. When any fail, report FAIL and surface the failed validations.