القائمة
Deep PR review that runs `code-review:code-review` and `pr-review-toolkit` agents in parallel, then hands off to `review-pr` for fix or post.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Show a sorted list of incomplete tasks from TODO.md.
Interactively walk through and address PR review comments one at a time. Accepts an optional PR number argument; defaults to the current branch's PR.
Review a pull request for issues and feedback.
Use when working with Jira or Confluence from command line, including authentication, searching issues with JQL, bulk operations, sprint reports, or creating/updating work items using acli
Update project documentation based on recent changes and current codebase state.
Create or append tasks to TODO.md in the current working directory.
| name | review-pr-ultra |
| description | Deep PR review that runs `code-review:code-review` and `pr-review-toolkit` agents in parallel, then hands off to `review-pr` for fix or post. |
| when_to_use | Use when the user says 'ultra review', 'review-pr-ultra', 'deep review this PR', 'thorough PR review', 'comprehensive PR review', 'multi-agent PR review', 'kitchen-sink review', 'ultra PR review', or any variation of wanting an exhaustive pull-request review that combines multiple reviewers. |
| argument-hint | [findings|fix|post] |
| disable-model-invocation | true |
Run two complementary upstream review skills in parallel, present their findings side by side, then hand off to review-pr for fix or post.
The two upstream skills cover non-overlapping ground:
code-review:code-review — git history, prior PR comments, CLAUDE.md compliance, confidence-filtered outputpr-review-toolkit agents — type design, comment accuracy, test coverage, silent failuresSame as review-pr:
findings — display side-by-side findings onlyfix — display findings, then run review-pr's Fix Flowpost — display findings, then run review-pr's Post FlowAskUserQuestion for fix / post / stopIdentify the target PR (default: current branch's PR). Dispatch both passes in a single assistant message so they run concurrently — sequential dispatch defeats the purpose, since tool calls must complete before the main turn continues.
Pass A — code-review:code-review
Launch a general-purpose subagent and tell it to invoke the /code-review:code-review slash command via the Skill tool (skill name: code-review:code-review), skip the final GitHub posting step, and return findings as structured data (file, line, severity, description).
Pass B — pr-review-toolkit agents (selective)
Dispatch the toolkit agents directly based on what the diff touches:
pr-review-toolkit:code-reviewerpr-review-toolkit:silent-failure-hunterpr-review-toolkit:pr-test-analyzerpr-review-toolkit:comment-analyzerpr-review-toolkit:type-design-analyzerEach agent must return findings as structured data and must not post to GitHub.
Show one section per source. Don't merge — the two skills use different severity scales and filtering, so cross-source agreement is for the user to weigh, not the orchestrator.
If neither pass returned anything actionable, say so and stop regardless of mode.
Behavior depends on the mode:
findings mode: stop here.fix mode: follow the Fix Flow defined in the review-pr skill.post mode: follow the Post Flow defined in the review-pr skill.AskUserQuestion with options fix, post, stop. Then run the corresponding flow (or stop).When tracking per-issue answers in Fix Flow, also include the source (which pass flagged the issue) so the user has that context.