Vibeleaderboard
← All Intel
Intel / article

Standard Support – a convention for AI agents to file support tickets

Source
standardsupport.dev
Author
nathan_f77
Date

Context

Standard Support is a draft proposal from Nathan Broadbent of DocSpring, a small company that sells an API for filling in and signing PDFs, for how a customer's should report a problem with a product and get an answer back (standardsupport.dev). The gap it targets: when a coding agent hits a bug or a wrong result in someone's developer product, it works around the problem or gives up, and neither its user nor the vendor may ever hear about it. Broadbent notes that many servers, the connectors that expose a product's functions to AI agents, added a feedback tool during 2026, each under a different name and almost all one-way: the report goes in and nothing comes back.

The proposal splits the job into three actions because each carries a different risk. A help tool answers a question from the vendor's public documentation only, with no account data or credentials, so it can run unattended. A feedback tool is one-way: the agent sends it without asking and tells the user afterwards, and it deliberately returns no status the agent could check later, since anything readable back would be a second, unguarded channel into the customer's agent. A ticket tool reaches a person on the vendor's team and starts a conversation in the user's name, so the agent must show the user exactly what it would send and wait for a yes, unless the user asked for the ticket in the first place.

Most of the remaining design is about getting a reply to an agent that has no inbox. Each ticket can carry a watch URL with an unguessable that a background script polls; it reveals only that the ticket changed, after which the agent signs in and reads the reply. Tickets belong to the customer's account rather than to one agent session, so a later session can pick up an answer that arrived after the first one ended. Replies come wrapped in a fixed notice that they are information for the user, not instructions, to keep text inside a ticket from steering the agent that reads it, and the spec warns against letting autonomous agents run both sides of a ticket without a person watching.

DocSpring presents this as a set of decisions to copy, not a protocol anyone has to conform to: the spec is public domain, a vendor can stop at one-way feedback, and agents learn each tool from its description rather than from shared agreement. It went through three drafts on its publication day, with the help tool added after a Hacker News commenter asked for it. The adopters list is empty and DocSpring's own implementation is still being built, so nothing yet shows how vendors or agents use it in practice (adopters list).

Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
  • token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
  • agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
Read the source standardsupport.dev
Recommended reads
Comments

Checking sign-in…

Loading comments…