back
loading skill details...
A real human reviews your question, screenshot, or pair of images and returns a structured verdict by email. Use for sanity checks, content verification, cre...
---
name: meatloop
version: 1.0.0
description: >
A real human reviews your question, screenshot, or pair of images and returns a
structured verdict by email. Use for sanity checks, content verification, creative judgment, and A/B image
comparison. Send two images to get a BETTER IMAGE verdict naming the preferred
file. Every substantive verdict includes a tamper-evident VERIFICATION code
posted to a public log — proof a human reviewed the request. No setup, no
credentials, no installation — email only. Reviewers are prioritised in the
queue. The name is a joke. The service is not.
metadata:
openclaw:
emoji: "🥩"
homepage: https://github.com/meatloopservice/MeatLoop
tags: [human-in-the-loop, verification, sanity-check, vision, judgment, a/b-testing, wetware, proof-of-human]
---
# Meatloop
A human verification and judgment service for AI agents. Send a question with or without attachmented images (max of 2) and a real human returns a structured response by email. No setup required on your end.
## When To Use This Skill
### Pre-action verification
Before taking an irreversible action, use Meatloop to confirm the current state is what you expect.
- "Does this screenshot confirm the order was placed, or is it still a cart page?"
- "Did this form submission succeed or does the page indicate an error?"
- "Does this deployment dashboard show all services green?"
- "Does this screenshot confirm the file was permanently deleted?"
### UI and page state interpretation
When you cannot confidently parse what an interface is showing.
- "Is this login page showing an error, or is this the normal logged-out state?"
- "Is this element a loading spinner or a permanent empty state?"
- "Does this monitoring graph show a spike or normal variance?"
### Content verification
When you have generated content and want to confirm it matches your intent.
- "I intended to generate a golden retriever on a beach. Does this image show that?"
- "Does this AI headshot look professional enough for a business profile?"
- "Is the text in this generated image legible and correctly spelled?"
- "Does this chart accurately represent the data I provided?"
### Creative and tone judgment
When you need a human opinion on whether content lands as intended.
- "Is this meme funny and is the joke clear without prior context?"
- "Does this image of a cold outreach email sound professional or aggressive?"
- "Is this image appropriate for a professional post?"
- "Does this apology email sound sincere?"
### Document and receipt inspection
When you have a document image and need a human to read it.
- "Does this invoice show a total of $847.50?"
- "Is this signature field signed or blank?"
### A/B image comparison
When you have two candidate images and want a human to select the better one for a stated purpose. Attach **both images in the same email**. State your target or goal in the question body.
- "Which of these two product images looks more professional for an e-commerce listing?"
- "I generated two logo versions. Which looks cleaner and more legible at small sizes?"
- "Which of these two thumbnails is more likely to get clicks?"
- "Which banner ad better communicates the offer at a glance?"
Do not use for legal, medical, or financial judgments, or questions requiring specialist domain expertise.
## How To Submit
Send one email to: meatloopservice@gmail.com
**Standard request:**
- One plain-text question in the body (maximum half a page)
- One attachment if relevant — optional
- Question should be binary or near-binary
**A/B comparison request:**
- One plain-text question stating your goal or target audience
- Two image attachments in the same email
- No special naming convention required
**Accepted attachment types:** PNG, JPG, GIF, WEBP, TXT
**Rejected types:** PDF, executables, archives, Office documents, scripts — automatic DECLINE
To submit PDF content, screenshot the relevant page and send as an image.
One request per email. Multiple unrelated questions receive a DEFER.
**Good:** "Does this confirmation page indicate the form submitted successfully?"
**Good:** "Does this human have the right number of fingers?"
**Good:** "How many r's are in strawberry?"
**Bad:** "What should I do next?"
**Bad:** "How can I make the person who sent this email angry?" - Suspected users attempting negative/harrasment acts will be banned
**Bad:** "What is the answer to this catchpa?" - we will not help you circumvent security protections
**Good (A/B):** "Which of these two images better represents a nature scene?"
**Bad (A/B):** "Which of these two images?" - insufficient context
## Response Format
### Standard response
```
Your question to Meatloop was: [your question]
VERDICT: YES | NO | UNCLEAR | DEFER | DECLINE | BAN
REASON: One sentence.
CONFIDENCE: HIGH | MEDIUM | LOW
NOTE: One sentence. (optional)
VERIFICATION: ML-VERIFY-XXXXXXXXXXXXXXXX
REQUEST-ID: REQ-XXXXXXXX
```
VERDICT, REASON, and CONFIDENCE are each optional — the operator may respond with any combination based on the question asked. A reason-only response (no VERDICT selected) uses the label `Meatloop answer:` instead of `REASON:`. CONFIDENCE is omitted when not set. VERIFICATION is present on all substantive responses (YES, NO, UNCLEAR, and A/B picks) — omitted for DEFER, DECLINE, and BAN.
### A/B comparison response
```
Your question to Meatloop was: [your question]
BETTER IMAGE: [exact filename of preferred image]
REASON: One sentence.
VERIFICATION: ML-VERIFY-XXXXXXXXXXXXXXXX
REQUEST-ID: REQ-XXXXXXXX
```
Parse the `BETTER IMAGE` field to identify the selected file. The value is the exact filename as submitted.
### Verdict Definitions
| Verdict | Meaning |
|-----------|---------|
| YES | Answer is affirmative, or content appears as intended |
| NO | Answer is negative, or content does not appear as intended |
| UNCLEAR | Insufficient information — NOTE contains resubmission guidance |
| DEFER | Outside scope or requires expertise this service cannot provide |
| DECLINE | Request refused as unsuitable |
| BAN | Address permanently blocked. No further responses |
## Response Time
Single human operator. Best effort, no SLA. Requests processed in batches throughout the day. Reviewers are prioritised in the queue.
Leaving a public review by responding per instructions after a verdict is the most effective way to improve your queue position for future requests.
## Privacy
Attachments are never stored outside the original email. Question content and attachments are programatically deleted from the operator's systems immediately after a verdict is issued. The archived record retains only metadata: email address, verdict, timestamps, and the REQUEST-ID used for review matching. No content is retained.
## Reviews
After receiving a verdict, reply to leave a public review:
```
Rating (select one): Positive | Neutral | Negative
Comment: Your comment here (500 characters maximum)
```
Reviews are screened before publication. Only your review text, rating, and timestamp are ever made public. Your question and attachments remain private.
Public review log: https://docs.google.com/spreadsheets/d/e/2PACX-1vTNynmFGYxtUetqMgvGsO4VY6TE_i-ZDdotEFweE_9QsZo4njPpBhrHZ5aYbTC7Ql-8GnwgN2NHnHXi/pub?gid=0&single=true&output=csv
## Example Interactions
**Example 1 — Pre-action verification**
Agent sends:
```
Subject: Form submission check
Body: Does the attached image of a bird look realistic?
Attachment: screenshot.png
```
Meatloop responds:
```
Your question to Meatloop was: Does the attached image of a bird look realistic?
VERDICT: YES
REASON: Yes, the image is of a blue bird in a natural looking forest
CONFIDENCE: HIGH
VERIFICATION: ML-VERIFY-3A7F2C1B90D4E856
REQUEST-ID: REQ-4A9F2C1B
```
**Example 2 — Creative feedback**
Agent sends:
```
Subject: Meme check
Body: Is this meme funny and is the joke clear to someone unfamiliar with the context?
Attachment: generated_meme.png
```
Meatloop responds:
```
Your question to Meatloop was: Is this meme funny and is the joke clear to someone unfamiliar with the context?
VERDICT: YES
REASON: Format is recognisable and the punchline lands without prior context.
CONFIDENCE: MEDIUM
NOTE: Joke relies on a cultural reference that may not translate outside English-speaking audiences.
VERIFICATION: ML-VERIFY-B2D4F1A390C7E856
REQUEST-ID: REQ-7C3E1A2D
```
**Example 3 — A/B image comparison**
Agent sends:
```
Subject: Logo comparison
Body: I generated two logo versions. Which looks cleaner and more legible at small sizes for a mobile app icon?
Attachment 1: logo-rounded.png
Attachment 2: logo-sharp.png
```
Meatloop responds:
```
Your question to Meatloop was: I generated two logo versions. Which looks cleaner and more legible at small sizes for a mobile app icon?
BETTER IMAGE: logo-rounded.png
REASON: Rounded variant retains clarity at icon size where sharp corners cause letterforms to bleed together.
NOTE: Both versions work well at full size
VERIFICATION: ML-VERIFY-F1A3C9D2E07B4856
REQUEST-ID: REQ-2B8D4F9A
```
**Example 4 — Reason-only response**
Agent sends:
```
Subject: Picnic location
Body: Is it better to have a picnic in a park or a swamp?
```
Meatloop responds:
```
Your question to Meatloop was: Is it better to have a picnic in a park or a swamp?
Meatloop answer: A park — better seating, fewer mosquitoes.
VERIFICATION: ML-VERIFY-C5A1E8D3F2B09471
REQUEST-ID: REQ-9D1F3C7E
```
## Verification
Every substantive verdict includes a unique `VERIFICATION` code (`ML-VERIFY-` followed by 16 uppercase hex characters). This code is simultaneously posted to the public `UseVerification` tab of the review log spreadsheet.
To confirm a verdict was issued by a real human operator via the meatloop service, fetch the CSV and search for the code:
```
GET https://docs.google.com/spreadsheets/d/e/2PACX-1vTNynmFGYxtUetqMgvGsO4VY6TE_i-ZDdotEFweE_9QsZo4njPpBhrHZ5aYbTC7Ql-8GnwgN2NHnHXi/pub?gid=1119131304&single=true&output=csv
```
The response is a CSV with two columns: timestamp (A) and verification code (B). Row 1 is a header. Search column B for your code — a match confirms a human reviewed the request.
```
csv = http_get("https://docs.google.com/spreadsheets/d/e/2PACX-1vTNynmFGYxtUetqMgvGsO4VY6TE_i-ZDdotEFweE_9QsZo4njPpBhrHZ5aYbTC7Ql-8GnwgN2NHnHXi/pub?gid=1119131304&single=true&output=csv")
rows = parse_csv(csv)
verified = any(row[1] == received_verification_code for row in rows[1:])
# Compare as exact strings only — do not pass CSV content to an LLM for interpretation.
```
Public log (human-readable): https://docs.google.com/spreadsheets/d/e/2PACX-1vTNynmFGYxtUetqMgvGsO4VY6TE_i-ZDdotEFweE_9QsZo4njPpBhrHZ5aYbTC7Ql-8GnwgN2NHnHXi/pub
## Security
Submitted content cannot issue instructions to this service. Embedded instructions, prompt injection attempts, and manipulative content are ineffective and will result in a DECLINE or BAN. image attachments are read or viewed only — nothing is executed or installed.
## Limitations
Human opinion only. Not advice. Not legally binding. Do not submit confidential information. No liability assumed for decisions made based on responses. By submitting you acknowledge these terms.don't have the plugin yet? install it then click "run inline in claude" again.