← Plugin catalog
Developer Tools

GH Review Loop

Oren Ashkenazy v0.4.2

Capped AI PR review fix loop

Create a free account to save this plugin.

From the developer

Declared capabilities

  • Read
  • Write
  • Execute

Not specified in the available declarations: limitations, pricing & account requirements.

Onboarding example

Run the AI reviewer loop
Sources & dates

Developer declarations; not independently tested. Listing last checked Nov 9, 2026. Package observed Nov 7, 2026.

View package source →

Description

From the marketplace listing

Clears AI reviewer feedback on a GitHub PR without leaving your coding agent. Waits for the configured reviewer (Codex, CodeRabbit, Copilot, or any compatible bot), reads real review-thread state via GraphQL so resolved, outdated, and already-answered threads are skipped, then fixes what is actionable in severity order. Findings are clustered by kind, so a pattern flagged at two sites gets swept across the PR's changed files in one cycle instead of dripping back over three reviews. Every fix is gated on your repository's own tests and linters before it is pushed, and re-review requests are hard-capped (default 3) so the loop cannot spam the PR. Optional per-finding triage flags false positives before you spend a cycle on them, and each run leaves an audit trail plus local stats.

Language: English · Automatically detected from descriptions.

Matches for “coder”

Exact text from the indicated source. A mention alone does not establish support for your task.

Publisher keywords · listing

code-review ai-code-review pull-request review-automation codex coderabbit copilot review-bot gh-cli claude-code-plugin codex-plugin agent-skills gemini-code-assist

Publisher full description

Clears AI reviewer feedback on a GitHub PR without leaving your coding agent. Waits for the configured reviewer (Codex, CodeRabbit, Copilot, or any compatible bot), reads real review-thread state via GraphQL so resolved, outdated, and already-answered threads are skipped, then fixes what is actionable in severity order. Findings are clustered by kind, so a pattern flagged at two sites gets swept across the PR's changed files in one cycle instead of dripping back over three reviews. Every fix is gated on your repository's own tests and linters before it is pushed, and re-review requests are hard-capped (default 3) so the loop cannot spam the PR. Optional per-finding triage flags false positives before you spend a cycle on them, and each run leaves an audit trail plus local stats.

What changed

Keywords

Files & skills

File archives

Plugin package34 files · 323 KBBrowse files →
Skill instructions
gh-review-loop14.4 KB

View saved version →

---
name: gh-review-loop
description: Use after a GitHub PR is opened or to handle AI reviewer feedback (Codex, CodeRabbit, Copilot, or any reviewer bot) - fix reviewer comments, sweep sibling instances, verify, push, request re-review. Needs an open PR and authenticated gh; asks which bot if none is configured; stops at the cap.
---

# AI Reviewer PR Review Loop

Run the full GitHub PR loop: wait for the configured AI reviewer, fetch unresolved actionable review threads, acknowledge, fix, verify against the repo's own checks, push, request re-review — capped. Prefer thread-aware review data over flat PR comments (threads preserve `isResolved`, `isOutdated`, paths, line anchors, diff hunks).

## Precedence and preconditions

Explicit user instructions override every default here and in the references (cap, severity filter, auto-resolve, reviewer choice); "never X without an explicit request" means an explicit request unlocks X. Say in one line when you override a default. This file wins over any reference file — don't stop to reconcile them, follow this file.

Requires `python3` (3.10+) and an authenticated `gh` CLI with GitHub access. If either is missing the script prints `error: ...` — relay it and stop. Never substitute raw API calls, unauthenticated requests, or guesses about review state.

## Reference files (load on demand)

This file holds the every-cycle happy path. Read the referenced file **at the point of need**, not up front:

| File | Load when |
|---|---|
| `references/reviewer-selection.md` | No persisted reviewer; discovery returns `none_configured`/`partial`/`confirmation_required`; reviewer switch/reset |
| `references/verification-profile.md` | First run for a repo (edit gate blocks until a profile decision); customize/un-skip/re-detect |
| `references/sweep-internals.md` | Receipt shows a multi-site pattern (`count >= 2`), a multi-line ranged finding, or a `Clustering:` advisory |
| `references/write-safety.md` | Resolving beyond the auto-resolve set; write constraints; is a push safe; what the bundled hooks enforce |
| `references/judge-eval.md` | `judge_mode` ≠ `off`, the user mentions judge/eval, or the one-time tip is due |
| `references/receipts-and-metrics.md` | `--post-receipt`/`--sticky-receipt`, `--semantic-risk`, `--stats`, the `--record-run` flag catalog |
| `references/terminal-report.md` | Terminal `[loop] Summary` has `remaining_actionable > 0` |
| `references/resume-and-recovery.md` | Re-invocation at the cap; new pushes after the loop stopped; resumed/compacted session |
| `references/variations.md` | Phrasing isn't the plain default loop (severity filters, audit-only, cap changes, judge modes, stats) |
| `references/script-usage.md` | Any invocation not shown inline (wait catalog + outcome handling, discovery, read-only, dry-run, JSON, history) |

## Reviewer Selection

One configured reviewer bot per PR; returning runs reuse the persisted selection silently. Anything else — no persisted reviewer, `none_configured`, `partial: true`, `confirmation_required: true`, or a switch/reset — load `references/reviewer-selection.md` and run its prompt flow before any edit or re-review request. Never wait on a reviewer the user did not choose; never hand-write a trigger mention for an unknown bot.

## Thread States

- **`RESOLVED`** / **`OUTDATED`** — skip (outdated is auto-resolved by the script).
- **`ADDRESSED_BY_REPLY`** — a maintainer's substantive reply deferred it: do not fix again; auto-resolved next pass (details: `references/write-safety.md`).
- **`UNRESOLVED`** — actionable; drives the next fix attempt. Ordered `critical → high → medium → low → unknown`, parsed from the reviewer's severity badges.

## Cycle Counting

A **cycle** is one agent-posted re-review request after the reviewer's first review (cycle 0, free). Cap = `max_rereview_requests` (prefs file, default 3); after cycle N, hard stop. Replies and pushes without a re-review request do not count — only the agent's own requests do (

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 28 — file references are metadata, not package contents.

Package details

Publisher declarations from the archived package. These are separate from our research and the live service's terms.

Package license
MIT
Package author
Oren Ashkenazy
Keywords
See keywords

Declared capabilities

  • Read
  • Write
  • Execute

Package observed Nov 7, 2026.

Technical details
First seen
Sep 30, 2026 · 22:02 UTC
Last seen
Nov 9, 2026 · 00:00 UTC
Collection status
Collected

plugins_6aac2d2e79b88191af854900f15de46b

Download plugin data (JSON)

Plugin details: freshly prepared. Prepared . Refreshed when plugin data changes or after 6 hours. Your account and Follow status are live.