メニュー
Review the branch's commits plus uncommitted work (the effective PR) with the aeview reviewer panel. Presents the review and stops; does not change code unless you explicitly ask.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
After you and the user have planned and implemented a change, loop with the aeview reviewer panel until it converges — running the project's gates, reviewing with the panel, and fixing the findings each cycle. Use when a change is already written and the user wants it reviewed and fixed until it passes the gates and the panel finds no new issues — "review until clean" or "loop until the panel is happy". It fixes what the panel finds; it does not build the change from scratch. For a one-shot review with no fixes, use the aeview skill instead.
Like aeview-loop — run after you and the user have planned and implemented a change — but it pauses every cycle to confirm each reviewer finding with the user before acting. For each finding it presents the issue, tailored resolution options (always including Ignore) and its recommendation via AskUserQuestion, then applies the chosen option. Use when you want to approve every fix the loop makes; for the autonomous version use aeview-loop.
Install the aeview CLI and its reviewer-panel skills on this machine, then verify the setup.
Review one or more specific commits with the aeview reviewer panel. Presents the review and stops; does not change code unless you explicitly ask.
Review a GitHub pull request with the aeview reviewer panel. Reviews the PR's diff and (by default) checks the PR out so reviewers read its real files. Presents the review and stops; does not change code unless you explicitly ask.
Review code changes with the aeview multi-harness reviewer panel. Use when the user asks to review a diff, branch, pull request, commits, or staged/working changes, wants a code review, or wants changes checked before merge or commit. You pick the scope from the request and may override reviewers and settings. Presents the review and stops; does not change code unless the user explicitly asks. For an implement-and-review loop, use the aeview-loop skill instead.
| name | aeview-effective-pr |
| description | Review the branch's commits plus uncommitted work (the effective PR) with the aeview reviewer panel. Presents the review and stops; does not change code unless you explicitly ask. |
| argument-hint | [--base <ref>] [--reviewers a,b] [aeview run flags] |
| disable-model-invocation | true |
Review the effective PR — the branch's commits plus any uncommitted work, against the
base — with the aeview panel. This is what your branch would look like as a PR right now.
Review-only — do not change code unless the user explicitly asks for a fix.
Raw arguments: $ARGUMENTS
--base <ref>: compare against <ref> → effective-pr:<ref>. If omitted, use bare
effective-pr (the base is auto-detected from the branch's open PR, else origin/HEAD /
main/master/trunk).--reviewers a,b, --allow-conflicts, …) forwards to aeview run.Always pass --json (the JSON gate is the reliable contract). A full panel takes a few minutes, so
run it as a background task rather than blocking:
aeview run --scope effective-pr[:<base>] --json [--reviewers …] [flags]
It prints its run id on stderr; let it finish (aeview status <run-id> --wait), then read the JSON
gate and present the review faithfully — verdict + each finding (severity, file:line, title,
recommendation). Do not editorialize or propose fixes. The exit code is the verdict: 0 approve ·
1 needs-attention · 2 error; full report via aeview result <run-id>.
After presenting the review, STOP. Do not fix anything on your own initiative — apply fixes only
if the user explicitly asks, or point them to the aeview-loop skill.