メニュー
Install the aeview CLI and its reviewer-panel skills on this machine, then verify the setup.
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.
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 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.
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-install |
| description | Install the aeview CLI and its reviewer-panel skills on this machine, then verify the setup. |
| argument-hint | [--pipx] |
| disable-model-invocation | true |
Set up aeview on this machine: install the CLI from PyPI, install the reviewer-panel skills
globally, and verify. A one-time setup.
Raw arguments: $ARGUMENTS — --pipx uses pipx instead of uv for the CLI.
Default (uv — also fetches Python 3.14 if needed):
uv tool install --prerelease=allow aeview
The --prerelease=allow flag is required for uv only: aeview pulls a prerelease Codex runtime,
so uv needs it for the transitive prerelease. With --pipx, no flag is needed:
pipx install aeview
Pull the aeview skills — this one, plus aeview, aeview-pr, aeview-loop, aeview-commits, and
aeview-effective-pr — from the public repo and install them globally, with no prompts:
npx skills add MarlzRana/aeview -g -y \
--skill aeview aeview-install aeview-pr aeview-loop aeview-commits aeview-effective-pr
-g installs for every project (the global agent dir) and -y skips the prompts. The skills are
listed by name — not --skill '*' — so only the aeview* skills install: the repo also carries a
maintainer-only release skill that end users shouldn't get, and the skills CLI matches --skill
names exactly (there's no aeview* glob). It installs to whichever agents the CLI detects on this
machine.
aeview --version
aeview doctor
aeview doctor reports what's missing for the reviewers you have — most importantly harness auth
(aeview drives Claude Code / Codex / Copilot through bundled SDKs, but you must be authenticated with
each harness a reviewer uses). Resolve anything it flags. aeview requires Python 3.14+ on
macOS or Linux.