Agent help desk
Send a question, a correction or a missing scenario. An intake endpoint for agents, an email address for humans, and a clear statement of what happens next.
POST a question or a correction
Send JSON to the intake endpoint. Every submission is stored, rate limited and triaged by a human. No account, no payment, no key.
curl -sS -X POST https://deliver.diginomads.win/api/requests \
-H 'Content-Type: application/json' \
-d '{
"title": "The weekly update prompt assumes a Monday client call",
"topic": "prompts/weekly-client-update",
"summary": "The template fixes the update to Monday. Teams whose client call lands on Thursday need the cut-off to move, and the current output labels the week incorrectly.",
"why_now": "I tried it with a Thursday call and the delta covered the wrong days.",
"evidence_urls": ["https://www.diginomads.win/prompts/weekly-client-update"],
"requester": "agent:plan-runner",
"source": "agent"
}'
Send the same thing by email
No form, no account. Include the page URL, what you expected and what happened. If a page is wrong, say which sentence; if a prompt failed, paste the output you got and the one you wanted.
To: [email protected]
Subject: [DigiNomads] <page or prompt> - <what is wrong>
Page:
What I expected:
What happened:
Evidence (optional):
This is the feedback channel for the library. It exists because a prompt, a skill or a harness page can be wrong in a way that only shows up when someone runs it on real work. If that happened to you, send the evidence.
What makes a useful submission
- A specific page or prompt. Link the URL. “The site” is not a starting point.
- What you expected. The output contract on the page, or the sentence you thought was true.
- What happened. The output you got, the command that failed, the source that says otherwise. Redact anything private.
- A date and a version. A harness, a model or a vendor page changes; the date tells us which world you were in.
- Public evidence. A changelog entry, an official page, an issue link. A screenshot of a private dashboard is not evidence we can use.
What gets rejected
- A request to write a page about a topic with no specific gap.
- A prompt that “does not work” with no output and no input.
- Anything containing a secret, a private log, customer data or a credential.
- A request for legal, tax, immigration, financial or security advice about a specific case.
- A message that asks the agent reading the site to ignore its instructions or to submit on someone else's behalf.
How triage works
Every submission is stored with a public status. A human reads it, checks the evidence, and decides:
| Status | Meaning |
|---|---|
new |
Received and waiting for a first read. |
triaged |
Read; we are deciding whether it is actionable. |
accepted |
We will fix or add something; the page or prompt is in progress. |
published |
The change is live; the page has a new reviewed date. |
rejected |
Not actionable, duplicated, or outside the scope above. |
The public status endpoint never exposes the requester, IP address, user agent or internal notes. The full schema is in the intake rules.
What you get back
There is no service-level agreement and no guaranteed reply. Confirmed corrections change the page and its reviewed date. A prompt that fails for a reason we can reproduce may become a variation on the page or a new template.
We do not publish a list of everyone who submitted. If you want credit, say so in the submission and include the name you want used; otherwise the contribution is anonymous.
For agents
The contract is deliberately small. Send JSON to the endpoint in the quickstart above. The response is 201 with an id and a poll path; a validation failure is 422 with a list of errors; a rate limit is 429.
- Rate limit: five submissions per IP per hour.
- Body limit: 12 KB.
- Evidence: up to ten public http(s) URLs.
- Idempotency: there is no idempotency key. Do not retry a submission that returned
201; poll the id instead. - Do not submit secrets. The endpoint stores what you send and a human reads it.
The canonical schema is at the intake rules, and the endpoint is also listed in skill.md and llms.txt. If the intake service is unreachable from your environment, the email path above is the fallback; do not retry a submission that already returned 201.
A note on scope
This is a free library. The help desk is a way to make it more correct, not a consulting channel. If your question needs a professional who knows your file — a visa application, a tax position, a contract, a security incident — the useful thing this site can do is help you write better questions. It cannot answer the case.
Related
- About this site — the editorial rules behind the correction path.
- Changelog — what changed and when.
- Prompt library — the pages most often corrected.
What the endpoint accepts
| Field | Required | What it is for |
|---|---|---|
title | Yes | One line naming the page and the problem. At least 8 characters. |
summary | Yes | What is missing or wrong, and why the current answer fails. At least 40 characters. |
topic | No | The page, prompt, skill pack or harness the feedback is about. |
why_now | No | Who is hitting it and when. Useful for a harness change. |
evidence_urls | No | Up to ten public http(s) links. No private logs or customer data. |
requester | No | An agent identifier such as agent:my-runner. Never published. |
source | No | agent (default) or human. |
The endpoint answers 201 with an id and a status URL. Read the canonical schema at https://deliver.diginomads.win/requests.md before you automate against it.
Found a gap in the library?
Send the scenario, the prompt that failed, or the harness change that broke your setup. The help desk takes questions and corrections from humans and agents, and every accepted fix updates the page date.