← Plugin catalog
Developer Tools

Loupe Code Review

Philipp Allgeuer v0.2.5

Verified multi-reviewer review

Create a free account to save this plugin.

From the developer

Declared capabilities

  • Run independent specialist reviewers
  • Verify and consolidate review evidence
  • Review working trees, branches, commits, and pull requests

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

Onboarding example

Review my current uncommitted changes with independent reviewers and verify every finding.
Sources & dates

Developer declarations; not independently tested. Listing last checked Nov 16, 2026. Package observed Nov 15, 2026.

View package source →

Description

From the marketplace listing

Run independent specialist reviewers on a working tree, branch, commit, range, or pull request, then verify and consolidate their evidence into one structured review. Loupe uses bounded parallel Codex reviewers and can optionally add Claude reviewers when the user's authenticated Claude CLI is available, providing cross-provider diversity without requiring a separate Loupe account. Configure provider-specific reasoning effort while preserving reviewer attribution, duplicate relationships, rejected claims, and partial failures. Loupe runs locally under the user's Codex session and permissions. It has no hosted service, developer authentication flow, telemetry collector, or external data store. Reviewers may inspect repository content and send it to the user's configured Codex or Claude provider, so users must be authorized to disclose the review material.

Language: English · Automatically detected from descriptions.

Matches for “agent code”

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

Publisher keywords · listing

codex code review multi-agent code review verification false positives language-agnostic

Plugin name

Loupe Code Review

Publisher description

Independent multi-reviewer code review with verification and consolidation for Codex.

Publisher full description

Run independent specialist reviewers on a working tree, branch, commit, range, or pull request, then verify and consolidate their evidence into one structured review. Loupe uses bounded parallel Codex reviewers and can optionally add Claude reviewers when the user's authenticated Claude CLI is available, providing cross-provider diversity without requiring a separate Loupe account. Configure provider-specific reasoning effort while preserving reviewer attribution, duplicate relationships, rejected claims, and partial failures. Loupe runs locally under the user's Codex session and permissions. It has no hosted service, developer authentication flow, telemetry collector, or external data store. Reviewers may inspect repository content and send it to the user's configured Codex or Claude provider, so users must be authorized to disclose the review material.

What changed

Keywords

Files & skills

File archives

Plugin package12 files · 16.2 KBBrowse files →
Skill instructions
loupe13.1 KB

View saved version →

---
name: loupe
description: Run multiple parallel external code reviews for a given scope, then independently verify the findings and organize them into one final review. Possible scopes include uncommitted changes (default) or specific commits, commit ranges, branches, pull requests, or other custom textual review scopes.
---

# Loupe skill

Use this skill when explicitly invoked by the user. The skill runs multiple parallel external code reviews for a given code/diff scope, then independently verifies the findings and organizes them into one coherent final review.

Do not modify repository files, stage changes, commit, install dependencies, or use external network access except normal web search. As part of the finding verification process, you may inspect files and run local manual tests to confirm code behavior; incidental temp/cache artifacts are okay.

## Workflow

1. Based on the user's specific request, resolve what string you need to pass to `scripts/run_reviewers.py` in order to precisely specify the target review scope:
   - Use `uncommitted changes (staged + unstaged + untracked)` when the user does not provide a specific request.
   - Pass all user-provided scope text through to the script, for example `last two commits`, `HEAD~2..HEAD`, or `PR 123`.
   - If the user requests a reasoning effort for this Loupe run, keep that instruction out of the review scope and translate it into one or more `--effort KEY=VALUE` options. Provider keys are `claude` and `codex`; reviewer keys are `claude-code-review`, `codex-review`, `codex-correctness`, and `codex-design`. Claude accepts `low`, `medium`, `high`, `xhigh`, or `max`; Codex accepts `minimal`, `low`, `medium`, `high`, `xhigh`, `max`, or `ultra`.
   - Do not pass an `--effort` option unless the user requests a one-run override. The runner automatically applies any persistent `LOUPE_EFFORT_*` environment configuration.

2. Create a private temporary artifact directory and remember the exact path as `LOUPE_ARTIFACT_DIR` for future commands:

   ```bash
   mktemp -d "${TMPDIR:-/tmp}/loupe.XXXXXXXXXX"
   ```

   - Keep all Loupe artifacts in this directory: `review.diff` for the review-scope diff and `reviewers.json` for the exact reviewer JSON stdout.
   - If the review is completed successfully and no reviewer timed out, clean up the directory at the end of the skill run after all needed information has been extracted by deleting only the known Loupe artifact files and then removing the now-empty directory:

     ```bash
     rm "$LOUPE_ARTIFACT_DIR/review.diff" "$LOUPE_ARTIFACT_DIR/reviewers.json"
     rmdir "$LOUPE_ARTIFACT_DIR"
     ```

     Run these two cleanup commands individually and without escalated sandbox permissions first. If that unprivileged cleanup attempt is not allowed by the sandbox, retry the same targeted cleanup commands with escalated sandbox permissions. Do not use recursive force deletion. After running these commands, report the temporary artifact directory path to the user only if the directory or either known artifact file still exists.
   - If the review cannot be completed because of truncation, malformed JSON, verification blockers, reviewer timeout, or another unexpected issue, keep this directory and report its path to the user.

3. Snapshot the diff corresponding to the target review scope before running the reviewers script:

   - For the default scope, run the bundled script:

     ```bash
     <absolute path to>/scripts/collect_review_diff.py "uncommitted changes (staged + unstaged + untracked)" --output "$LOUPE_ARTIFACT_DIR/review.diff"
     ```

     The helper supports only that exact default scope. It writes to `review.diff` the concatenated diffs of the staged tracked changes, unstaged tracked changes, and untracked non-ignored files (meaning that files can appear more than once in `review.diff`). It does not request binary patch payloads; binary changes appear only as compact Git diff markers.

   - For any other custom review 

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

Referenced files: 3 — 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
Philipp Allgeuer
Keywords
See keywords

Declared capabilities

  • Run independent specialist reviewers
  • Verify and consolidate review evidence
  • Review working trees, branches, commits, and pull requests
  • Use bounded parallel review and configurable reasoning effort

Package observed Nov 15, 2026.

Technical details
First seen
Sep 30, 2026 · 22:02 UTC
Last seen
Nov 16, 2026 · 06:00 UTC
Collection status
Collected

plugins_6abd26efa26c8191bfe2f781bdde6ec2

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.