ThomasMoreAI/legal-skills-open / au/general/skills/legal-writing-mblauberg
legal-writing-mblauberg
VerifiedUse for drafting, reviewing, condensing, or source-checking Australian legal documents (forms, affidavits, submissions and orders), legal decision overviews, or correspondence. Not for academic, engineering, or general prose; use its specialist.
references/argument-structure-and-paragraphing.md
View on GitHub# Argument Structure And Paragraphing
How to build a submission so each paragraph makes one point, nothing is re-argued, and a
judicial reader gets question-answer-reasons at a glance. Load with
`legal-concision-and-anti-ai.md` for any advocacy or submission work. This reinforces the
SKILL Concision Pass; it does not replace it. The boundary rules at the end govern: never
trade an honest legal negative for elegance.
## Write For The Decision-Maker
- The judgment-ready standard: draft submissions capable, if the tribunal agrees, of forming
the substance of the reasons. That one test drives accuracy, restraint, issue-by-issue
structure and citation discipline (Heydon: sufficient authority, not too much). Written
submissions are the first half of the appeal; a bad start is hard to recover.
- Write for a busy generalist meeting the matter cold: assume less time, less record
knowledge and less niche law than the drafter has. Explain and orient; never assume
familiarity. The greatest sin is a reader who feels lost and must re-read.
- If the case turns on a statute, contract or order, get to its words fast. Long background
before the operative text reads as avoidance - the suspicion is the wording is against you.
Put the provision inside the point it governs, not in an early block the reader must flip
back to.
- File or attach a chronology where the forum allows it; a dated, sourced chronology keeps
narrative detail out of the argument body.
## Front-Load The Whole Document
- A fresh reader can state the relief sought, the grounds and why within about 90 seconds.
The opening delivers the upshot before any detail.
- Deep issue: frame the central question as a syllogism of about 75 words or fewer, in
separate sentences - legal premise, then concrete factual premise, then the question -
concrete enough that a non-lawyer follows it. Put the premises before the question so the
answer is implicit. One issue block, not a string of abstract "whether" clauses. State the
issue concretely: embed the decisive fact ("whether the discretion miscarried where both
eyewitnesses to the collision had died"), never "whether the tribunal erred", which tells
the reader nothing.
- Make any summary-of-argument detachable: it reads alone as a complete persuasive document,
issue by issue - your position in a sentence or two, the flaw in the other side's position,
then the fuller why. Ideally it makes reading the rest optional.
- An arresting, concrete opening beats a recital: "There are four fundamental flaws in the
appellant's contentions", then the flaws numbered, each footnoted to the record, is a
worked gold standard. State the case as you would to an interested but distracted
colleague; no jury speech.
- Opening order of operations, by scope. These front-loading devices compete for the same
opening; pick by unit size, do not stack them. Whole document: upshot paragraph (relief +
grounds + why), then the deep issue, then a detachable summary where length warrants one.
Single ground or short section inside a larger document: open with the one-sentence
ground-answer carrying the embedded decisive fact; use a deep-issue block only where the
ground is long enough to need it; relief lives in the document's orders block, not in the
section's first paragraph.
## Headings Carry The Argument
- Each argument-bearing heading is a one-sentence assertion of that section's conclusion, not
a topic label. The heading list, read alone top to bottom, states the whole case.
- Test: extract every heading into a bare list. If a line is a neutral tag ("Background", "The
completion date") rather than a claim ("The reasons disclose no finding on the agreed date"),
it fails.
- Exception: genuinely structural sections - a procedural history, a shared-framework or
umbrella block - may stay labels. Do not force a contention where the section only orients.
## One Point Per Paragraph
- Open every paragraph with its contention - a contestable position the reader could agree or
disagree with - not a question or a recital. Conclusion first (CRAC/CREAC).
- One developed point per paragraph. If the honest one-line caption needs "X and (separately)
Y", where Y could stand or fall alone, split into two. Tie-breaker against Synthesise
Fact-Clusters below: instances that prove the same proposition merge into one paragraph;
propositions that could each independently carry a ground or sub-point split.
- Close each ground or section on its conclusion in one or two sentences. Do not trail off on a
citation or a fresh sub-point.
- In a rule paragraph, funnel: the broadest governing principle first, then the narrower
qualifications and exceptions; order multiple authorities highest court to lowest.
- Put the most decisive sentence of a unit at its start or end (the emphatic positions); keep
routine chronology in the interior.
## Synthesise Fact-Clusters Into One Proposition
- When several facts, messages, transactions or authorities each make the same small point, do
not give each its own narrated sentence or paragraph with the conclusion repeated after every
item. State the pattern and its single conclusion once as the paragraph's topic sentence, then
hang the instances beneath it as a compressed anchor list (a semicolon series or lettered
sub-paras), each keeping its own pinpoint.
- Test: does this paragraph exist only to add one more instance of a pattern the paragraph above
already states? If yes, fold it in as a clause plus pinpoint, not a new paragraph.
- Draw the conclusion the cluster proves - "on his own contemporaneous words the sum had no
settled existence" - rather than leaving the reader to add the instances up. A synthesised
proposition with the facts marshalled under it reads stronger and shorter than a fact-by-fact
recitation. Synthesise WITHIN one issue; never fuse distinct legal questions (jurisdiction,
reasons, procedural fairness) into one "errors" block to save space - that destroys the
separate-CRAC structure and makes each ground unwinnable.
- Embed the decisive fact in the issue itself: "whether the decision-maker
denied procedural fairness by relying on a report disclosed after the hearing
without inviting a response", not "whether procedural fairness was denied".
One concrete issue line does the marshalling several
abstract paragraphs attempt.
## State It Once; The One Allowed Echo
- Front-load once: state each shared premise, rule, figure and chronology once,
in the framework or umbrella block, and have later sections cite it by name.
See the SKILL concision rule that one document does one job; a project-specific
document structure may define the stable cross-reference nodes.
- Emphasis is not redundancy. Deliberate restatement is legitimate only at a structural
primacy point - the opening frame and one close - and only as reinforcement that carries new
evidence or stakes. Every other repeat is redundancy: cut it to a cross-reference.
- The reason is credibility, not only length: a repetitious filing is skimmed, and a reader who
sees the same point three times discounts the writer. Cutting repetition is a persuasion move,
not housekeeping.
- Protective formulas are rationed the same way. A recurring shield formula (a convergence
theory, a context-only renunciation, a purpose-limit on sensitive material) appears in full at
its defined home, its first grounds-body use and the per-ground application close; every other
deployment folds to a bare node cite (`(C.1)`, `(C.7)`). By the closing parts a repeated formula
reads as a tic, and repeated disclaiming draws the eye to the very material it disclaims. Two
limits: no deployment site is ever left without at least the bare cite (the shield must survive
everywhere it protects), and the one site that actually deploys the sensitive material keeps the
full formula - reducing the shield at the point of use is a cut too far. Distinguish the
renunciation echo (reduce) from a located evidentiary purpose-limit on a specific source (keep).
- Pack coherence (multi-document filings): when the filing is a pack - submissions plus an outline,
schedules and a draft orders sheet - one point still has one home and every other document
cross-refers to it. A cross-reference must name the target document and a stable locus (its
heading, paragraph or schedule number, not a page number that re-flows on rebuild), and it must
resolve to something that exists. The same figure, defined term, date or label must read
identically in every document in the pack; reconcile them before filing,
because a reader who finds the same defined term or total stated two ways
discounts both.
## Select And Aim
- Grounds-selection: run the strongest grounds, not every available one. A stack of marginal
grounds signals under-analysis and dilutes the strong points. Appellate judges say this
plainly: seven to ten grounds raises a presumption none has merit (Aldisert, adopted by
McHugh J); the scattergun approach "obscures the true issues" and "camouflages the
pleader's best points" (Ipp JA).
- Ground-tied diagnostic: every submission ties to a specific ground or issue. If a
submission cannot be articulated cleanly and supported by record references and legal
principle, the defect is usually in the ground itself - amend it or abandon it; do not
paper over it with prose.
- One argument is one argument. Do not dress a single contention as a "primary limb" and an
"alternative limb": false alternatives read as under-confidence and invite the tribunal to
treat the point as weaker than it is. Reserve express alternatives for genuinely distinct
legal routes to the relief.
- Do not double-number. If sub-sections are already numbered (E.2, E.3 ...), do not also label
them "Limb 1", "Limb 2": two enumeration schemes for one structure read as machine scaffolding.
Let the numbered headings carry the count and refer to the points in plain ordinals ("first ...
secondly ... thirdly"). Reserve "limb" for a genuine term of art (the limbs of a two-part test,
the division-of-assets limb of a binary); for enumerated strands under an already-numbered
structure prefer a plain word ("aspect", "point"). Client-facing jargon such as "limb" or
"strand" where a plainer word does the same job is an avoidable readability cost.
- Favourable-direction filter: every retained passage must move the tribunal toward the relief
sought. If you cannot say with conviction why a passage helps, cut it.
- Primacy: lead with the strongest ground, and within a multi-limb ground lead with the
strongest limb. Order threshold-first, then strongest-to-weakest - a potentially dispositive
jurisdiction/nullity ground leads regardless of when it arose in the hearing.
- Prefer a small set of distinct, supportable grounds. Three or four is an
editing heuristic, not a procedural limit **(skill default)**. Do not merge
legally distinct grounds merely to hit it; do fold a genuinely overlapping
complaint into its stronger owner.
- Cut versus relocate - keep the two distinct. Cut a genuinely weak point outright (a hedged
"in the alternative" makeweight dilutes the strong grounds and signals they needed padding).
But do NOT cut strong, cited analysis or evidence to hit a page target:
relocate genuinely schedular detail with a pinpoint cross-reference. Cutting substance guts the evidentiary
spine; relocating keeps the force and saves the page. Sentence-level trimming alone floors at a
few pages - a large page cut comes from homing repeated analysis and relocating cited working,
not from shortening sentences.
- Schedule-role discipline: a schedule - of authorities, of a calculation, of a reasons-deficiency
- is a reference and record device, not a place to argue. The argument lives in the submissions;
the schedule is the marshalled backup a submission points to at its point of use, cited by number
where it does its work. Do not smuggle a contention into a schedule that the submissions do not
make. Where several submissions in a wave draw on the same material, consolidate it into one
standalone schedule they each cite rather than repeating it in each - the one-home rule applied to
the pack's reference layer.
- Tables are a readability device, not a page-saver. Converting reasoned prose to a multi-column
table almost never shortens the document and usually lengthens it: the cell text that preserves
the argument (inference plus pinpoint) is paragraph-length, and paragraph-length text in a narrow
column wraps heavily and carries cell padding, so it consumes more vertical space than the same
words flowing full-width. Reserve tables for genuinely tabular material - short-cell data, a
figure-to-source map, a compact matrix a time-pressed reader parses at a glance. If you table to
aid comprehension, do it knowing it costs a page or two, not to save them. Real page savings come
from prose synthesis (fold repeated instances into one proposition plus an anchor list) and
relocation, not from re-laying prose into a grid. Measure page count on the actual render before
claiming a table saved space.
- Density ceiling: a page limit is a ceiling, not a target, and saturation is its own defect. When
a draft nears the forum's limit, the trigger is structural, not sentence-level: re-run
grounds-selection (cut a weak ground outright), re-home repeated analysis to its one owner, and
relocate cited working to a schedule - sentence trimming alone floors at a few pages. Length
erodes persuasion before it breaches a rule: the longer the document, the less of it is read,
and a reader who suspects padding skims. Density means each retained page carries argument the
relief needs; saturation means every available page is full.
- Theory of the case, held honestly: a single organising theme gives the document its spine, but
never bend, select or soften found facts to preserve the theme. If the facts as found do not fit
the theory, attack the fact-finding directly or change the theory.
## Adverse Authority And Candour
- Verify and comply with the current professional-conduct, candour and forum
duties concerning adverse authority. Where disclosure is required, it is not
a tactical choice; identify the authority accurately and meet it fairly.
- Distinguish on the facts first: argue the facts of this case sit outside the adverse
precedent's reasoning. Attack the correctness of an adverse decision only as a last resort,
and never attack an authority's worth or the opponent's motives - attack applicability and
reasoning.
- Confront the obvious weakness up front and in your own words. Volunteering and answering the
weak point projects command of the case; side-stepping it persuades the reader there is no
answer. Concede a genuinely non-essential point cleanly - candour is a weapon, and a
located concession buys credibility for every contested point that follows.
- Anticipate, never attribute. An expected answer is framed as expected ("the
other party may answer that ..."), sourced to the record where it has
actually been said, and never put into
the opponent's mouth as a position the record does not contain (see the integrity rules in
`source-boundary-and-citations.md`).
- Do not take an isolated sentence of the decision below out of context: identify the impugned
passage precisely (verbatim quote or tight summary, always with the record pinpoint) and meet
it as it was meant. A reader forced to ask "where do we find that?" has stopped being
persuaded.
- Explain, do not label: bald characterisations of the other side's argument ("hopeless",
"nonsense", "misconceived" without more) do none of the tribunal's work. State why the
argument fails - the premise it needs, and the source that removes that premise.
## Reply Submissions
- A reply answers; it does not re-argue. Take the opponent's distinct points in a sensible
order, state acceptance or rejection of each, and give the short reason with its pinpoint.
Do not restate the case-in-chief - cross-refer to it by paragraph.
- Answer the argument actually put, at its strongest reading, not a weaker paraphrase. Meeting
a straw man concedes the real point silently.
- Silence on a point in reply reads as concession: sweep the opponent's document once, list
every distinct contention, and account for each - answered, adopted, or expressly not
pressed and why.
- Hold the whole argument in writing: a point saved for the hearing that has no trace in the
written submissions carries "an unattractive air of afterthought" and may not be reached at
all on the papers.
- Anticipate; never attribute. A primary submission may answer points of principle the opponent
is expected to press, but only framed expressly as anticipation ("X advances, and Y
anticipates, the following points"), stated neutrally at their strongest, and answered from
what the record already carries. Never attribute to the opponent an argument no document of
theirs yet makes. Hold predicted-only answers - points with no present record footprint - for
the authorised reply round: filing them early previews the reply case and reads as arguing
with yourself.
- Fold answers into the deciding node; do not build a standalone rebuttal Part. Where the
opponent has actually served a point, answer it inside the ground or node that decides it - a
cite to his filed paragraph plus a one-line answer - not in a separate "answers to the points
against us" section. A standalone anticipation/rebuttal Part duplicates the grounds (inviting a
state-once breach), and, where a reply round is already timetabled, reads as seeking a leave to
reply you already have. Reserve a separate section only for genuinely new matter that has no
home in a ground.
## Sentence Economy
- Average around 20 words. Treat any sentence over about 24 words, or carrying three or more
list items, as a split-or-tabulate prompt - not a hard rule where precision or a pinpoint
genuinely needs the length (the canonical statement is in `australian-english-house-style.md`).
- Three or more parallel quantifications, versions or events, each carrying its own source and
outcome, are a table, not a prose chain (columns on the pattern figure / where advanced /
basis / fate). Byte-preserve every figure, quote and anchor in the conversion, and leave the
table unnumbered if nothing cites it, so it never disturbs an existing table series.
- Vary sentence length deliberately; uniform length reads as machine rhythm. Land the decisive
point in the shortest sentence, after a longer reasoned one.
- Active voice with strong verbs naming the actor; cut nominalisations and agentless passives.
- Plain English: cut `herein`, `said` (as adjective), `pursuant to`, `inter alia`, `the same` (as
pronoun); replace wordy connectives (`in order to` -> `to`; `due to the fact that` -> `because`;
`in the event that` -> `if`; `with regard to` -> `about`).
- Cite sparingly: one pinpoint for an uncontested point; reserve full treatment for the central
contested ground; do not string several cases for a settled proposition.
## A Deliberate Fresh-Reader Pass
- Before finalising, read the document cold as the busy judicial reader. Can you recover relief,
grounds and reasons from the headings and the first sentence of each paragraph alone? Where
not, that section fails front-loading - repair it before polishing.
## Boundary (Carry From Concision)
- None of these rules strips an honest legal negative or source boundary - non-service,
non-admission, under-appeal, a reserved issue. Keep one genuine-boundary qualifier per
proposition. Positive form, em-dash count 0, contentions not findings, anchors never cut to
shorten.
## Sources
- Australian judicial doctrine: Michael McHugh, "The Essence of Appellate Advocacy" (ABA 2012:
concrete issues, best point first, argumentative headings, proposition-before-support,
distinguish-first); Susan Glazebrook, "Effective written submissions" (2014: detachable
summary, candour as a weapon, edit-ruthlessly, page limit is a ceiling); M J Beazley,
"Appellate Advocacy" (NSW Bar CPD 2015: submissions-relate-to-grounds diagnostic, explain
don't assert, the four-flaws gold standard); J D Heydon, "Reciprocal Duties of Bench and Bar"
(2007) 81 ALJ 23 (judgment-ready standard, whole argument in writing, restraint); Pepper/Jagot,
"A Basic Guide to Good Written and Oral Advocacy" (adverse-authority duty, write-the-decision
test); M D Kirby, "Ten Rules of Appellate Advocacy" (1995) 69 ALJ 964.
- Brief-writing canon: Bryan A. Garner, "The Deep Issue" (Scribes Journal of Legal Writing,
vol 5) and *The Winning Brief*; Scalia & Garner, *Making Your Case*; Ross Guberman, *Point
Made* (synthesis, topic sentences, issue framing); Ruggero J. Aldisert, *Winning on Appeal*
(ground selection, do not drown good arguments); Columbia Law School Writing Center,
"IRAC/CRAC/CREAC" and "Headings and Umbrella Sections".
- Empirical: Long & Christensen (2008) 45 Idaho L Rev 171 (frequent intensifiers correlate with
losing appeals); Black, Hall, Owens & Ringsmuth (2016) 4 J Law & Courts 377 (emotional
language costs merits votes); Flammer (2010) 16 J Legal Writing Inst 183 (judges prefer plain
English about 2:1); Long & Christensen (2011) 12 J App Prac & Process 145 (readability alone
does not win appeals - argument strength does); Spencer & Feldman (2018) 22 Legal Writing 61
(readability correlates with first-instance success).
- Distilled further in 2026-07-01 and 2026-07-02 multi-source research passes (session
transcripts; conclusions encoded here).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |