aeview-loop

name	aeview-loop
description	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.
argument-hint	[--reviewers a,b] [--no-commit] [--max-cycles n] [--post-comments] [what to build or fix]

aeview-loop

You and the user have already planned and implemented the change. This skill loops with the aeview reviewer panel until it converges — you run the gates, run the panel, triage the findings, and fix them. Unlike the aeview skill (review-only), this skill changes code; but the change is already written before you start, so here you fix what the panel finds rather than build from scratch.

Raw arguments: $ARGUMENTS

• Freeform text is context for the review — the scope/files to focus on, or a note on what the change does.
• --reviewers a,b and other aeview run flags pass straight through to the panel.
• --no-commit — don't commit between cycles; work stays in the tree (changes the review scope).
• --max-cycles <n> — override the default cap of 5 review cycles (the loop stops as soon as it converges regardless). The user can also name this in plain language ("at most 3 cycles").
• --post-comments — post each cycle's review onto the PR and reply on every finding's thread as an audit trail. Forces --scope pr, needs an open PR (creates one if missing), and rejects --no-commit.

Run the loop

Follow one loop, matching the flags — the two are complete and self-contained, so follow only the one that applies:

• Default → references/loop.md.
• With --post-comments → references/loop-post-comments.md (posts the review to the PR and works the findings as PR threads).

Required summary

When the loop ends, give the user a summary:

• Addressed — each finding you fixed, and what changed.
• Ignored — each finding you didn't, with the reason.
• The cycle count and the outcome (converged, or stopped at the cap with N open).