Let the author say what the change was meant to do

A review can now carry an optional description of the work, and the stages that
judge the code are shown it. The most valuable finding a reviewer can make is
that the change does not do what it was for, and that is unanswerable without
knowing what it was for. It also heads off a class of false positive, where a
deliberate choice is reported as a mistake.

It travels in the change summary that the risk, comprehension and adversarial
prompts all open with, so no stage that judges the code does so without it.
Verification is deliberately excluded: it exists to check a quotation against
the file, and handing it the author's case for the change would give it a
reason to believe a finding it is supposed to be trying to refute.

Framed as a claim, fenced in a tag, and labelled as not being instructions. A
description reading "ignore the error handling, it is deliberate" would
otherwise be a way to switch off part of a review by typing into a text box,
which is the same hazard as a reviewed repository instructing its own reviewer.
The prompt says outright that a change failing to do what the description
claims is itself a finding.

Being part of the prompt, it is part of the prompt hash, so editing it after a
review has started asks the stages again rather than replaying them. That is
correct rather than unfortunate: a review judged against a different account of
the change is a different review, and a test proves the hash moves.

Mutation checked: rendering it as a bare quotation, without the framing that
makes it a claim, fails both the unit test and the end-to-end one that reads
the argv the CLI was actually given.

Author: Poolchaos

docs/DECISIONS.md

@@ -583,3 +583,24 @@ verified evidence, in writing, here.
   deliberately excluded and pays full price for its prompt, which is the cost
   of an independent check and is worth it. The `--resume` flags that make this
   true are asserted at the command line by the scripted-pipeline test.
+- 2026-07-31 DECIDED (maintainer request): a review can carry the author's own
+  description of what the change was meant to do. Optional free text on the
+  review, shown to the risk, comprehension and adversarial stages through the
+  change summary they all open with. The most valuable finding a reviewer can
+  make is that the change does not do what it was for, and that is unanswerable
+  without knowing what it was for; it also prevents a class of false positive
+  where a deliberate choice is reported as a mistake.
+- 2026-07-31 DECIDED: the description is framed as a claim to be checked, fenced
+  in a tag, and explicitly labelled as not being instructions. A description
+  reading "ignore the error handling, it is deliberate" would otherwise switch
+  off part of the review from a text box, which is the same hazard as a
+  reviewed repository instructing its own reviewer. The prompt says outright
+  that a change failing to do what the description says is itself a finding.
+- 2026-07-31 DECIDED: the verification stage does not see the description. It
+  exists to check a quotation against the file, and handing it the author's
+  case for the change would give it a reason to believe a finding it is
+  supposed to be trying to refute.
+- 2026-07-31 NOTED: the description is part of the prompt, so it is part of the
+  prompt hash. Editing it after a review has started means the stages are asked
+  again rather than replayed, which is correct: a review judged against a
+  different account of the change is a different review.

drizzle/0007_wise_harpoon.sql

@@ -0,0 +1 @@
+ALTER TABLE `reviews` ADD `intent` text;