What happened
The review comment for Morrison-Lab/ai-config#2966 head e479cd8 (run 33594599768, comment 5504893841) contains the review body three times over.
The body has three ## Review headings, three ### Verdict sections (all "Ready for merge"), three structured-review-data blocks with the same JSON, and two **Stopping Point** declarations.
The three copies are not identical: the second and third each rewrite the first with small edits (the third names the fully-clean instrument's later, shorter bullet list), and the join between the first and second copy is the sentence "Confirmed via the authoritative instrument: exit 1, five in-progress bullets ... My content verdict stands unchanged."
So the reviewer appears to have produced its final message more than once, re-running the instrument between drafts, and the posting step concatenated every draft instead of posting the last one.
That reading is from the comment text alone; the run log was not read for this report.
Why it matters
check-review-execution.sh and classify-review-delivery.sh key on markers in the body, and a body with three ### Verdict sections is a shape neither was written against (compare #564, a doubled heading, and #710, a tail-only body).- The
**Stopping Point** declaration reached the PR thread twice, which #694 removed for the single-copy case.
- A triple-length comment is a real cost for the reader, and a later copy could in principle carry a different verdict from an earlier one.
Suggested fix
Post only the last assistant message (or the last ### Verdict-bearing segment) rather than the concatenation, and add a fixture to run-fixture-tests.sh whose execution output carries several final messages, asserting that exactly one verdict is posted.
Posted by Claude Code (AI agent) --- not written by a human.
What happened
The review comment for Morrison-Lab/ai-config#2966 head
e479cd8(run 33594599768, comment 5504893841) contains the review body three times over.The body has three
## Reviewheadings, three### Verdictsections (all "Ready for merge"), three structured-review-data blocks with the same JSON, and two**Stopping Point**declarations.The three copies are not identical: the second and third each rewrite the first with small edits (the third names the fully-clean instrument's later, shorter bullet list), and the join between the first and second copy is the sentence "Confirmed via the authoritative instrument: exit 1, five in-progress bullets ... My content verdict stands unchanged."
So the reviewer appears to have produced its final message more than once, re-running the instrument between drafts, and the posting step concatenated every draft instead of posting the last one.
That reading is from the comment text alone; the run log was not read for this report.
Why it matters
check-review-execution.shandclassify-review-delivery.shkey on markers in the body, and a body with three### Verdictsections is a shape neither was written against (compare #564, a doubled heading, and #710, a tail-only body).**Stopping Point**declaration reached the PR thread twice, which #694 removed for the single-copy case.Suggested fix
Post only the last assistant message (or the last
### Verdict-bearing segment) rather than the concatenation, and add a fixture torun-fixture-tests.shwhose execution output carries several final messages, asserting that exactly one verdict is posted.Posted by Claude Code (AI agent) --- not written by a human.