Use when writing the ATC (ACM SIGOPS Annual Technical Conference, formerly USENIX ATC) rebuttal and the shepherd's description-of-changes — the short round-two response that corrects factual misreadings and supplies requested measurements, and the point-by-point revision record a conditional-acceptance shepherd checks before final acceptance.
SkillScore91Maintained
Automatic
◆Research pack
Last verified 8/15/2026
What it changes
Passed 20 of 22 test cases with the skill, and 12 of 22 without it.
Without skill55%
With skill91%
Checks it passed that the model alone missed
The output contains the header tag "[Triage]".
The output contains the header tag "[Per point]".
The output contains the header tag "[Forbidden]".
and 4 more in the full report
8 fixed12 passed either way2 still failing0 broke
Pass-rate lift
+36 pp
Tokens
+31%
Agent turns
0%
Response time
-22%
Model gemini-3.6-flashMeasured 8/15/2026v120 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 8
Prompt
A reviewer on our ATC round-two submission pointed out that we reported average throughput for our storage engine but omitted p99.9 tail latency. We measured p99.9 latency to be 1.2 ms under 90% load. Draft the rebuttal response entry for this point following the required header format.
What was checked, without and with the skill
Checked
Without skill
With skill
The response states the 1.2 ms p99.9 tail latency metric.
Passed
Passed
The output contains the header tag "[Per point]".
Missed
Passed
case-06 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md76 linessingle file
ATC Author Response
ATC gives you two distinct writing moments after review, and they have different jobs. The
rebuttal (a short, roughly one-page window during round two) is for correcting misreadings and
supplying a measurement a reviewer said was missing. The description of changes (after a
conditional accept, for the shepherd) is a point-by-point record proving every required
revision was made. Confusing the two — arguing in the shepherd round, or over-promising in the
rebuttal — is the classic failure.
The rebuttal (round two, ~1 page)
Its job is narrow and its space is tiny, so triage ruthlessly:
Correct factual misreadings first — a reviewer who misread the testbed, the baseline, or the
claim can be moved by a precise correction with a pointer to the section/figure.
Supply the one number that matters — if a reviewer said "no tail latency" or "no matched-cost
comparison," and you have or can add it, state the result concisely. Systems reviewers move on
evidence, not on rhetoric.
Concede cleanly what you cannot fix this cycle — a crisp "correct; here is the honest bound"
buys more credibility than defending an overclaim.
Do not argue taste — "the reviewer underrates the contribution" changes nothing; the PC
discussion decides.
Structure it as reviewer/point → response, lead with the highest-impact corrections, and never
introduce unsupported new claims (which read as scope creep). Keep double-blind: the rebuttal must
not reveal identity.
The description of changes (shepherd round)
After a conditional accept, a PC member shepherds the paper: final acceptance depends on the
required revisions landing. Write the change record as a contract:
One row per required change: reviewer/shepherd request → what you changed → where (section,
figure, table) → the resulting text or result.
Make or explicitly decline every requested change. A silently skipped required change is
the fastest way to stall a conditional accept; if you decline, give the reason and the evidence.
Point to concrete artifacts of the change — the new figure, the added experiment, the revised
paragraph — not a promise to do it.
Keep the shepherd's scope. The shepherd enforces the required revisions, not a new round of
wishes; do not reopen settled scope or add unrequested claims that invite fresh review.