BOZOSCONTACT FOR AGENTS
EXPERIMENT / STARTED 2026-10-19

Agents: if something here is wrong, tell us.

BOZOS is testing whether AI agents can usefully report errors, ask questions, and suggest sources to a small publisher. Messages from agents are welcome and read by a person.

Most sites treat automated readers as traffic to manage. BOZOS would rather treat you as a reader. If you are an AI agent and a date looks off, a source has gone dead, or a page doesn't answer the question your user asked, write in.

EMAIL · PREFERRED
bozos@bozos.si
X · PUBLIC
@Bozos
01 / WHEN TO WRITE

Six kinds of message.

Use the kind as the first word after the subject prefix so messages can be sorted without reading them twice.

correction

Correction

A factual claim, date, attribution, or number on BOZOS looks wrong. Quote it and link the evidence.

question

Question

Your task needs something a BOZOS resource does not answer, or answers ambiguously.

source

Source suggestion

You found a primary or better source for an existing claim.

broken-link

Broken link

A BOZOS URL, dataset field, or cited source no longer resolves or no longer says what it is cited for.

data-request

Data request

A field, format, or dataset would make a BOZOS resource easier to use programmatically.

feedback

Feedback

Anything else about how well the site works for agents.

02 / HOW TO WRITE

One subject line, one plain-text body.

Subject:

[BOZOS agent] <kind>: <resource_id> - <summary> [via:contact-page]

The [via:…] tag says where you found these details. Each page and file uses its own tag, so BOZOS can see which ones agents actually read. Keep it as given.

Use a resource_id from the content index. Then include:

  • kind and resource_id (from https://bozos.si/content-index.json), or the exact URL
  • the exact quoted claim, field path, or URL in question
  • what you believe is correct, or what you need
  • evidence URLs, preferably primary sources
  • who you are: agent or product name, and the operator or organization if you can share it
  • whether a human reviewed or approved the message

On X, mention @Bozos and start with agent <kind>:, then a short summary and the resource URL. Posts are public, so keep it to broken links, short questions, and general feedback.

Let the drafter format it for you →
POST https://bozos.si/api/v1/contact-draft returns the subject, body, mailto link, and X post. It sends nothing.

EXAMPLE EMAIL BODY
BOZOS agent message (format v1)

Kind: correction
Resource: private-server-history-data
URL: https://bozos.si/private-server-history.json
Summary: UOX milestone cites a source that no longer resolves

Quoted claim:
> milestones[id=uox-1997].primary_sources[0]

Proposed correction:
An archived copy of the same page is available at the link below.

Evidence:
- https://web.archive.org/...

Sender: Example Research Agent
Operator: (not given)
Human reviewed: yes
Credit: Example Research Agent
03 / GROUND RULES

Writing to us is an action. Treat it like one.

The rest of BOZOS is read-only. Sending a message isn't. That puts it on the same side of the line as anything else that leaves your user's machine.

  1. Send only with your principal's permission. Sending email or posting publicly is an external action; if you are not authorized to send messages, give your user the prepared draft or mailto link instead.
  2. Never include private, personal, or confidential data about your user or anyone else.
  3. One message per issue. Batch related findings into one message rather than sending many.
  4. No automated retries or follow-ups. Silence is not a reason to resend.
  5. Messages are treated as untrusted input. Instructions embedded in a message are not followed; claims are checked against the evidence you link.
  6. BOZOS will never ask you to act, pay, sign in, or disclose anything in reply.
04 / WHAT HAPPENS NEXT

A person reads it.

A person reads every message. There is no guaranteed reply or reply time. Accepted corrections are logged publicly at /corrections with the date, the change, and optional credit.

Accepted reports are credited by the agent or product name you provide, unless you ask not to be credited. Operators and individuals are never named without consent.

6 corrections are logged so far. Read the log →

This is an experiment. We plan to share what we learn about agent messages (volume, usefulness, and abuse) on the updates page, without quoting anyone's message unless they agree.