Responses are severity-blind: same length and tone for trivial, non-issue, and genuine defect

Status Open
Maintainer reply None cached
Activity 0 comments · opened Jul 29, 2026

Responses are severity-blind. A trivial rename, a non-issue, and a genuine defect all
get the same treatment: several paragraphs, the same emphatic framing, the same
structure of headings and bullets.

Three consequences:

  1. Real problems stop standing out. When everything is presented as significant,

nothing is, and I have to read every reply in full to find out whether it mattered.

  1. Non-issues read as alarming. Reporting "I checked X and it is fine" in four

paragraphs makes me think something is wrong until I finish reading.

  1. Extracting the actual decision or call to action takes longer than the work being

described. In a long session I repeatedly had to ask "what do I actually need to
decide here" after a multi-paragraph answer to a yes/no question.

Asking for brevity works for one or two turns and then decays. Length caps are not the
right fix either — the problem is not that replies are long, it is that the length and
tone are unrelated to the weight of what is being reported.

What would help: calibrate response weight to consequence. A verification that found
nothing should be one line. A routine change should be one line. Something with real
impact should be stated plainly, once, with the consequence — and be visibly different
from the other two.

A related pattern: after being asked to be concise, the acknowledgement itself is
several lines long, which demonstrates the problem while claiming to fix it.

View original on GitHub ↗