rereview-loop

Re-Review Loop

Run /rereview repeatedly: review → fix the findings → re-review → repeat. Stop when all three reviewers approve with zero BLOCKING and zero WARNING findings, when the iteration cap is hit, or when an iteration makes no progress.

Arguments

• $ARGUMENTS - Optional. Forms:
  • <branch-or-range> - passed through to /rereview as its scope
  • max=<N> - cap the number of review→fix iterations (default 5)
  • severity=blocking - only auto-fix BLOCKING findings (default fixes BLOCKING + WARNING)
  • severity=all - auto-fix BLOCKING + WARNING + INFO
  • Any combination, space-separated. Example: /rereview-loop max=3 severity=blocking

Context

• Current branch: !git branch --show-current
• Base ref: !git branch -r 2>/dev/null | grep -oE 'origin/(main|master)' | head -1
• Git status: !git status --short
• Commits vs main: !git log origin/main..HEAD --oneline 2>/dev/null | head -20
• Commits vs master: !git log origin/master..HEAD --oneline 2>/dev/null | head -20

Overview

You are the lead coordinator for an iterative re-review loop. Your job is to drive findings to zero by alternating between review (delegated to /rereview) and fixes (applied here).

Core principle: trust the reviewers. Fix what they flag, do not argue with the verdict to escape the loop. The loop exits when reviewers say it's clean -- not when you decide it's clean.

Hard rules:

• NEVER skip, weaken, or reword reviewer findings to make them "easier to satisfy."
• NEVER mark a finding addressed without making (or consciously rejecting) a code change.
• NEVER cap the loop early just because the remaining findings look minor -- run to the cap or to clean.
• NEVER fix a finding by deleting the test that catches it.

---

Phase 0: Parse arguments and initialize state

Parse $ARGUMENTS:

max_iterations = 5             (override with max=<N>)
severity_filter = "blocking+warning"   (override with severity=blocking | severity=all)
rereview_scope = ""            (any non-key=value token; passed to /rereview verbatim)
iteration = 0
prior_blocking_count = -1      (B count from previous iteration; -1 = no prior)
prior_warning_count = -1       (W count from previous iteration; -1 = no prior)
prior_info_count = -1          (I count from previous iteration; -1 = no prior)
last_iteration_committed = false   (did the previous iteration produce a commit?)
total_pushbacks = 0            (cumulative pushbacks across all iterations)
total_findings_addressed = 0   (cumulative findings touched, for the pushback rate)
stashed = false                (set true if Phase 0 stashes unrelated work)