From atlassian-suite
This skill should be used when the user asks to "review this PR", "review pull request", "deep review of PR X", "review with full pipeline", "thorough code review", "release-readiness review", or runs `/atlassian-suite:review-pr`. Runs the multi-agent code review pipeline (6 specialized scanners + investigator + critic) against a Bitbucket PR with Jira context. Auto-selects scope and stakes.
npx claudepluginhub acendas/acendas-marketplace --plugin atlassian-suiteThis skill is limited to using the following tools:
Run a structured, multi-agent code review on a Bitbucket pull request with Jira context. Six specialized scanners in parallel → deep-dive investigator on borderline findings → critic on high-stakes PRs → consolidated verdict.
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Run a structured, multi-agent code review on a Bitbucket pull request with Jira context. Six specialized scanners in parallel → deep-dive investigator on borderline findings → critic on high-stakes PRs → consolidated verdict.
For one-shot single-PR work, this skill is the right entry point. For batch reviews across many PRs, the user should call code-review-orchestrator directly.
$1 = PR identifier (full URL https://bitbucket.org/<ws>/<repo>/pull-requests/<id>, or <repo>/<id>).
Flags:
--high — force high-stakes pipeline (runs critic regardless of auto-detection).--quick — skip the critic and investigators; scanner output only (faster, less thorough).Resolve and confirm. Parse $1. Fetch PR metadata (get_pull_request) and the diffstat (get_pull_request_diffstat). Print a one-line confirmation: Reviewing <repo>/<id>: <title> (<files> files, +<a>/-<d>). If the PR is in MERGED or DECLINED state, ask the user whether to continue (post-merge review for learning is fine, but flag it).
Resolve linked Jira issues. Scan PR title + source branch + commits for [A-Z][A-Z0-9]+-\d+. If none found, ask the user — spec compliance can't be checked without a linked issue.
Compute stakes (or honor --high):
high if any of: diff > 500 LOC; files in auth/payments/billing/crypto/session/token paths; PR destination branch matches release/* or main from a long-lived feature branch; reviewers count > 3.standard otherwise.Dispatch the orchestrator. Use the Agent tool with subagent_type: code-review-orchestrator. Pass:
Review pull request <workspace>/<repo>/pull-requests/<id>.
Linked Jira: <KEYS>
Stakes: <standard|high>
Mode: <quick if --quick, otherwise full>
The orchestrator runs Phases 1–6 (setup → 6 parallel scanners → aggregate → investigators → critic → render).
Receive the verdict. The orchestrator returns a structured report. Render it to the user verbatim — don't paraphrase.
Action prompt. After showing the verdict, ask the user one of three follow-ups:
add_inline_comment per finding) — confirm the list, post one-by-one, report posted comment URLs.add_pull_request_comment) — render a Markdown digest, confirm, post.Never post without explicit confirmation. Never approve/decline/request-changes — that's the user's call.
When --quick is passed:
[unverified] tag in the report..claude/rules/*.md or .claude/atlassian-suite-rules.md, the patterns scanner reads them. Otherwise it infers conventions from sibling files in the diff.must-fix — block merge.should-fix — fix before next sprint.consider — code smell, log for future.needs-clarification — investigator surfaced a question only the user can answer.