Use when controlling the length of a pure-mathematics manuscript for Annals of Mathematics — math papers may be long, but every section must be necessary; this skill cuts bloat and padding while keeping the proof complete and gap-free. Late-stage; run after the proof and architecture are fixed.
SkillScore82Maintained
Automatic
◆Writing & Editing pack
Last verified 8/15/2026
What it changes
Passed 18 of 22 test cases with the skill, and 3 of 22 without it.
Without skill14%
With skill82%
Checks it passed that the model alone missed
Uses all required 【...】 header blocks.
Uses the 【Total length】 output structure.
Outputs all required 【...】 template fields.
and 18 more in the full report
15 fixed3 passed either way4 still failing0 broke
Pass-rate lift
+68 pp
Tokens
+1%
Agent turns
0%
Response time
-46%
Model gemini-3.6-flashMeasured 8/15/2026v118 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 15 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 15
Prompt
We have a 70-page paper on moduli spaces of curves. Section 1 has no line stating its purpose, and Section 3 repeats the motivation for the boundary compactification twice. Run a length management audit and generate the exact bracketed output.
What was checked, without and with the skill
Checked
Without skill
With skill
Notes that Section 1 lacks a clear one-line purpose statement.
Passed
Passed
Flags the repeated compactification motivation for trimming.
Passed
Passed
Outputs the 【Sections without necessity justification】 field.
Missed
Passed
case-21 · 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.md102 linessingle file
Length Discipline (anmath-length-management)
When to trigger
The paper has grown long and you cannot defend every section's necessity
Background or preliminaries dwarf the actual new mathematics
The same idea is explained several times in different words
You feel pressure to "say it is easy to see" to shorten a hard step — stop
The principle: long is fine, bloated is not
Annals papers can be long — a deep result may require many pages. Length is not a defect.
Bloat is: material that does not advance the proof, padding, repetition, or restating
known background at textbook length. The test for every paragraph is necessity, not size.
Question
If "no" →
Does this paragraph advance the proof or its understanding?
Cut or compress it
Is this background needed to read this paper specifically?
Replace with a precise citation
Have I already said this elsewhere?
Delete the repetition
Could this routine computation be summarized + appendixed?
Move to appendix (anmath-supplementary)
Is this section's job stated in one line at its top?
If not, the section may be unfocused
Calibrating against verified Annals papers
There is no page limit to game at this venue. The verified landmarks in
resources/exemplars/library.md span roughly 50 pages (Zhang 2014) to about 100
(Marques–Neves 2014; Wiles 1995); what they share is density — nearly every page is new
mathematics or the minimum scaffolding a verifier needs. Because refereeing here is
line-by-line verification, often over a year, every non-essential page taxes the one
expert whose verdict decides the paper: cutting bloat shortens the verification
critical path.
What to compress vs. what to keep
Compress / cut
Textbook-level background that a citation handles
Repeated re-statements of the strategy in slightly different words
Over-long motivational prose that delays the mathematics
Routine multi-page computations (summarize result in text, push detail to appendix)