chatbot

A rule that is off by default: it reports text a chat assistant addressed to its user: an offer, flattery, a knowledge disclaimer once a decisions table turns it on.

What it catches

Text a chat assistant addressed to its user: an offer, flattery, a knowledge disclaimer.

Dimension

stance: the text reads as a verdict on a person, or as addressed to someone else. A finding is counted on the dimension of its rule, the cost it names to the reader.

Why

"I hope this helps" and "great question" answer a chat the reader never saw. Zero hits over the docs pages, changelog fragments and merged pull requests; the letter patterns ("Dear", "I am writing to") judge only a change description, which is never a letter. Measured 2026-10-10 before a fix: 88.9% precision on its cases (8 of 9), firing on 0.3% of human review comments (1.2% of their replies), 4.2% of bot replies and 0.1% of AIDev pull requests; 29 of 30 human review-comment firings read were a person answering in the thread ("You're right, ...", "feel free to", "great catch"). A review reply's reader saw the chat, so there an agreement, an invitation, an offer and an answer's opener are left alone; "I hope this helps" and the disclaimers still count. Measured 2026-10-11 over the adjudicated labeling round: 30.8% precision (16 of 52) before, 100% (14 of 14) after. In a reply, "you're absolutely right", "great question" and "I'm sorry I" were wrong in all 21 labeled firings, so they count only on a page or in a description, and words quoted from an issue or a comment are left alone. It stays off until 30 labeled firings show it.

Default decision

The code is PRF4004. The decision depends on the kind of text judged:

Changing it

A decisions table sets this rule to deny, advise or off, for everything judged or for the files a glob matches. Proofread reads the table from the file its -decisions flag names:

{
  "rules": {"chatbot": "deny"},
  "paths": {"blog/**": {"chatbot": "off"}}
}

rules applies to every file. A glob in paths applies to the files it matches, and the last match wins. A rule set to off never reports.

Seeing it

A finding names the rule, its code and the decision that applies, and links here:

{"rule": "chatbot", "code": "PRF4004", "decision": "off", "url": "https://eli.gladman.cc/magus/reference/proofread/chatbot/"}

proofread rules prints this entry with every other rule's.

See also

• All proofread rules - every rule, grouped by what it checks
• Writing rules - how the rules apply to this repository's pages and pull requests
• The findings contract - the fields every finding carries