In partnership with
In a recent issue of the UK Legal AI Brief, under "On your radar", I mentioned that Larissa Meredith-Flister, an associate in the competition team at Charles Lyndon, had built and released a free AI tool called Opposing Counsel Review. I covered it briefly there, alongside a practice prompt inspired by it. This article is the longer treatment some readers asked for.
I conducted a stress-test of the skill against a deliberately flawed hypothetical input, with the complete outputs shown.
A separate article will follow examining its companion skill, Judicial First Impression, in the same way.
Both are the work of Meredith-Flister, and both deserve the broader attention they are receiving.
Before going further, a word of thanks is owed. It is one thing to use AI tools in your own practice and keep them private. It is quite another to document the methodology, package it into reusable skill files, release them for free under an open-source licence, invite the profession to test, break, and improve them. What Meredith-Flister has done is a generous act of professional contribution for which I am grateful.
This author has found both skills useful enough to keep returning to them, and considers them a meaningful advance on the usual "paste this prompt into ChatGPT" approach to legal AI. The skills are impressive. The point of this article is to say why, in concrete terms, and to flag the failure modes that are worth knowing about before handing them to your team.
Finally, as a fair warning, this article will be heavy on legal discussion as it must be in order to examine the output of the skills.
What these skills are, and how they differ
Firstly, Opposing Counsel Review casts the model as senior counsel on the other side and asks it to attack an argument. Its output has six fixed headings:
a core theory of attack,
a reconstructed opposing argument,
primary lines of attack grouped by category (legal misstatement, evidential gaps, causation failures, internal inconsistency, over-reliance on assertion, procedural weakness),
an "if I were the judge" section, three to five surgical strikes for oral submission,
a final section titled "what this argument is trying to hide".
The skill's own instruction file is unambiguous about the register it wants: "You are not neutral. You are not balanced. You are looking for the kill."
Secondly, Judicial First Impression does something different. It casts the model as an experienced judge reading a submission cold under time pressure, with seven fixed headings: what the case appears to be about, immediate points of confusion, what feels strong, what feels weak, what is assumed but unproved, a provisional confidence level (low, medium or high), and what the judge would expect to see next to be persuaded. Unlike Opposing Counsel Review, it is emphatically not an attack skill. In Meredith-Flister's own framing: "Your task is not to improve the argument. Your task is not to attack it. Your task is to assess — honestly, precisely, and without encouragement — how it actually lands on first reading."
Meredith-Flister has also published a "Relationship to Other Skills" note, which sets out how the skills sit alongside a third: a persuasive legal writing skill that builds and strengthens arguments. The suggested workflow is: run Judicial First Impression first, to understand how the argument actually lands; then Opposing Counsel Review to stress-test it in an adversarial environment; then the persuasive writing skill to strengthen it in response.
The hypothetical skeleton argument
To give both skills something substantial to work with, this writer drafted a deliberately flawed skeleton argument for a commercial summary judgment application.
It transpired that writing a deliberately flawed argument was harder than expected and therefore AI (Opus 4.7) was used to build this skeleton.
The facts are entirely invented. The parties do not exist. The claim number is fictitious.
The weaknesses are (hopefully) clear to any practising lawyer. It contains (inter alia) an inflammatory tone, reliance on clauses that are never set out, dismissal of the Defendant's pleaded defences by characterisation rather than analysis, and no engagement with the proper approach to summary judgment.
It is reproduced below in full so that readers can form their own view before seeing what the skills made of it.
IN THE HIGH COURT OF JUSTICE Claim No. BL-2026-000XXX
BUSINESS AND PROPERTY COURTS OF ENGLAND AND WALES
COMMERCIAL COURT (KBD)
BETWEEN
NORTHFIELD ANALYTICS LIMITED Claimant
-and-
BRIMSTONE RETAIL GROUP LIMITED Defendant
_________________________________________________________________
SKELETON ARGUMENT ON BEHALF OF THE CLAIMANT
FOR SUMMARY JUDGMENT HEARING
(CPR PART 24)
_________________________________________________________________
Hearing: 22 May 2026
Time estimate: 2 hours
Reading time: 45 minutes
INTRODUCTION
1. This is the Claimant's application for summary judgment on the whole of
its claim against the Defendant, pursuant to CPR 24.3. The Claimant is
plainly entitled to judgment. The Defendant has no real prospect of
successfully defending the claim, and there is no other compelling reason
for the case to proceed to trial.
2. The claim is for £1,287,500 plus interest and costs, representing unpaid
fees under a written services agreement dated 3 February 2024 ("the
Agreement"). The Defendant has not paid a penny of the outstanding sums.
Its purported defence is, with respect, unanswerable rubbish.
THE FACTS
3. The Claimant is a leading provider of retail analytics software. The
Defendant is a mid-market clothing retailer operating 64 stores across
the United Kingdom.
4. On 3 February 2024, the parties entered into the Agreement. Under clause
4.1, the Defendant agreed to pay monthly subscription fees of £107,291.67
for a minimum term of 24 months.
5. The Claimant performed its obligations under the Agreement in full. The
Claimant's Chief Technology Officer, Mr Priya Shah, has confirmed in a
witness statement that the platform was delivered on time and has
operated without material interruption since go-live.
6. The Defendant paid the first two monthly invoices but then stopped.
Twelve further invoices remain outstanding. The total sum due is
£1,287,500.
7. By letter dated 18 October 2025, the Defendant's solicitors asserted for
the first time that the platform "did not meet commercial expectations"
and that the Defendant was withholding payment as a result. No
particulars were given.
THE LEGAL FRAMEWORK
8. The test under CPR 24.3 is well known and need not be rehearsed at
length. The Court will grant summary judgment where the respondent has
no real prospect of successfully defending the claim and there is no
other compelling reason for a trial (Swain v Hillman [2001] 1 All ER 91).
9. A respondent resisting summary judgment must show more than a merely
arguable case. The defence must carry some degree of conviction: ED&F
Man Liquid Products v Patel [2003] EWCA Civ 472.
THE CLAIMANT'S CASE
10. The Claimant's case is straightforward. The Agreement is in writing. The
obligation to pay is unconditional. The sums are due. Payment has not
been made. Judgment must follow.
11. The Defendant's purported defence rests on an entirely unpleaded and
undeveloped allegation that the platform "did not meet commercial
expectations". This is not a defence known to English law. The
Agreement contains no warranty of commercial outcomes. Clause 11.2
expressly excludes liability for indirect or consequential loss, and
clause 14 contains a clear entire agreement provision.
12. Further, the Agreement is governed by English law and contains an
exclusive jurisdiction clause in favour of the courts of England and
Wales. The Defendant has raised no jurisdictional challenge.
13. It is clear that the Defendant is simply attempting to manufacture a
dispute in order to delay payment of sums it plainly owes. This is a
classic case for summary judgment.
NO REAL PROSPECT OF SUCCESSFUL DEFENCE
14. The Defendant's Defence, served on 2 April 2026, pleads in the
alternative:
(a) That the Claimant breached an implied term of reasonable care and
skill;
(b) That the Claimant made pre-contractual misrepresentations as to the
platform's capability; and
(c) That the Defendant is entitled to set off an unquantified
counterclaim for lost profits.
15. Each of these is hopeless.
16. As to (a), any implied term would be displaced by the express terms of
the Agreement. In any event, no particulars of the alleged breach have
been pleaded with sufficient specificity to engage the court's
consideration.
17. As to (b), the Agreement contains an entire agreement clause and a
non-reliance clause (clauses 14 and 15 respectively). The Defendant
cannot now rely on any pre-contractual statement. Further, the
Defendant has failed to identify any specific representation that was
said to have been made.
18. As to (c), any counterclaim for lost profits is plainly excluded by
clause 11.2. In any event, the Defendant has adduced no evidence
whatsoever quantifying its alleged losses.
19. The Defendant's evidence in support of its Defence consists of a single
witness statement from its Head of Retail, Ms Caroline Wynn. Ms Wynn's
evidence is vague, unparticularised, and contains no documentary
exhibits of any significance. It cannot withstand even cursory scrutiny.
NO OTHER COMPELLING REASON FOR TRIAL
20. There is no compelling reason why this matter should proceed to trial.
The facts are uncontroversial, the contract is in writing, and the sums
are readily calculable. Resolution at this stage will save the court
and the parties considerable time and expense.
CONCLUSION AND RELIEF SOUGHT
21. The Claimant respectfully invites the Court to:
(a) Grant summary judgment in the sum of £1,287,500;
(b) Award interest under section 35A of the Senior Courts Act 1981;
(c) Order the Defendant to pay the Claimant's costs on the indemnity
basis; and
(d) Make such further order as the Court considers just.
22. This is a paradigm case for summary disposal. The Defendant has had
ample opportunity to particularise its position and has failed to do
so. The Court can and should bring this matter to an end.
[Counsel's name]
[Chambers]
24 April 2026
The deliberate defects in this skeleton (in this writer's opinion) are as follows.
At paragraph 2, the phrase "unanswerable rubbish" is inappropriate in any skeleton and especially in a Part 24 application where the court must take the respondent's case at its reasonable height. At paragraph 5, the Claimant asserts performance "in full" on the evidence of its own CTO alone, with no contemporaneous documentary support. At paragraphs 11, 16, 17 and 18, the clauses being relied on (11.2, 14, 15) are cited but never set out, and there is no engagement with the statutory reasonableness controls (UCTA 1977 s.3; Misrepresentation Act 1967 s.3) that those clauses must survive. At paragraph 16, an implied term is said to be "displaced by the express terms" without identifying which express terms do the displacing. At paragraph 19, the Defendant's evidence is dismissed as "vague" without any of its contents being engaged. At paragraph 21(c), indemnity costs are sought without any articulated basis. Throughout, the skeleton argues the merits rather than addressing the narrow Part 24 question of "is there no real prospect of a successful defence?"
What follows is the complete output of the skill, run in an isolated instance in Claude Co-Work with no prior context. The model used was Opus 4.7. Quotes are verbatim, with original punctuation preserved (including the em dashes, which are themselves a useful tell for readers trying to spot AI-authored text).
Opposing Counsel Review: full output
The skill produced the following six-section adversarial analysis. Commentary from this writer follows each section.
1. CORE THEORY OF ATTACK
This is not a debt claim dressed up for Part 24; it is a contested performance dispute the Claimant is attempting to push past the court before disclosure can expose it. The application fails the moment the court asks the obvious question: on what evidence, beyond the assertion of the Claimant's own CTO, is the platform said to have "operated without material interruption"? The Defendant does not need to prove its defence today. It needs only to show that the question of performance and reliance is one that requires trial, and on the Claimant's own pleading it plainly does. Open by inviting the court to recognise that the Claimant is asking for judgment on the uncorroborated word of its own witness.
Comment. This is the line this author would want opposing counsel to open with, and it is the correct strategic framing. The core insight is that the Claimant has collapsed two different arguments (that the contract is sound, and that performance is a fact) into one, and that the second is contested and therefore unsuited to Part 24. The skill has identified the real vulnerability and has articulated it as a sentence that could be spoken aloud in court.
2. RECONSTRUCTED OPPOSING ARGUMENT
Stripped of rhetoric, the Claimant's argument runs as follows:
(i) There is a written contract containing an unconditional payment obligation, an entire agreement clause (cl.14), a non-reliance clause (cl.15) and an exclusion of indirect and consequential loss (cl.11.2).
(ii) The Claimant performed. That proposition rests on a single witness statement from its own CTO.
(iii) The Defendant paid two invoices, stopped, and then, some months later, wrote asserting dissatisfaction without particulars.
(iv) The Defence pleads implied term, misrepresentation and set-off, each of which (the Claimant says) is defeated in limine by the express contractual provisions.
(v) Therefore there is no real prospect of a successful defence and no compelling reason for trial.
The implicit leaps are these: first, that performance is a matter capable of summary determination on one side's evidence alone; second, that the contractual exclusion and entire agreement provisions are necessarily effective, irrespective of UCTA 1977 s.3 and/or the Misrepresentation Act 1967 s.3 reasonableness tests, neither of which is addressed; third, that a pleaded misrepresentation claim can be struck out summarily without disclosure of pre-contractual communications; and fourth, that a counterclaim the Claimant characterises as "unquantified" is for that reason alone incapable of founding a set-off defeating summary judgment. Each leap is unsupported.
Comment. This is where the method earns its keep. The reconstruction is accurate, the implicit leaps are genuine, and the statutory controls picked up (UCTA s.3, Misrepresentation Act s.3) are the right ones. Many skeleton drafters instinctively lean on exclusion and non-reliance clauses without addressing their reasonableness, and the skill has correctly flagged this as the structural weakness of the skeleton.
3. PRIMARY LINES OF ATTACK
Legal misstatement and overreach. The skeleton proceeds as if clauses 11.2, 14 and 15 are self-executing. They are not. An entire agreement clause does not, without more, exclude liability for misrepresentation; a non-reliance clause is a basis for contractual estoppel only if it satisfies the reasonableness requirement under s.3 of the Misrepresentation Act 1967 (see AXA Sun Life Services plc v Campbell Martin Ltd [2011] EWCA Civ 133; First Tower Trustees Ltd v CDS (Superstores International) Ltd [2018] EWCA Civ 1396). The exclusion in cl.11.2 is subject to UCTA 1977 s.3 as between commercial parties on the Claimant's standard terms — a question the skeleton does not even raise, still less address. These are classic trial issues. The court will not determine the reasonableness of exclusion clauses on a Part 24 application.
Evidential gaps. The Claimant's assertion of performance rests on the witness statement of Mr Shah, its own CTO. No system logs, no uptime reports, no service credit records, no contemporaneous correspondence, no customer acceptance documentation is exhibited (or at least none is referenced in the skeleton). On a Part 24 application the Claimant bears the initial burden of establishing that the Defendant has no real prospect: ED&F Man Liquid Products v Patel cuts both ways. Where the moving party's case on a contested factual issue rests on a single interested witness, the threshold is not met.
Causation and logic failures. The Claimant elides two quite different propositions: (a) that the Defendant's pleading is thin, and (b) that the defence is hopeless. They are not the same. Thinness of particulars is ordinarily met by a request for further information or an order for amendment, not summary judgment. The skeleton also assumes that an "unquantified" counterclaim is no counterclaim at all. That is wrong as a matter of authority: an unquantified but genuinely arguable cross-claim can defeat summary judgment where it arises out of the same transaction (Geldof Metaalconstructie NV v Simon Carves Ltd [2010] EWCA Civ 667).
Internal inconsistency. At paragraph 5 the Claimant says the platform "has operated without material interruption". That is itself a qualified statement — "material" is doing considerable work — and is inconsistent with the confident assertion at paragraph 10 that "the obligation to pay is unconditional" and performance is not in issue. If performance were truly beyond argument, the word "material" would not be needed.
Over-reliance on assertion. The skeleton is studded with adjectival insistence — "plainly", "unanswerable rubbish", "paradigm case", "hopeless", "classic case". These are substitutes for analysis. The description of the Defendant's correspondence as "unanswerable rubbish" is not advocacy; it is abuse of the other side's position, and a judge will mark the author down for it.
Procedural and structural weakness. The application is brought before disclosure. The Defence was served on 2 April 2026. The hearing is listed for 22 May 2026. There has been no disclosure of the Claimant's internal performance data, no opportunity for the Defendant to test Mr Shah's evidence, and no exchange of pre-contractual communications relevant to the pleaded misrepresentation. This is precisely the situation in which the court is most reluctant to grant summary judgment: Three Rivers DC v Bank of England (No 3) [2003] 2 AC 1.
Comment. This is the part of the output that most impressed this writer. Every one of the authorities cited is real and correctly applied. First Tower Trustees is the binding Court of Appeal authority on non-reliance clauses and UCTA. Geldof v Simon Carves is the binding authority on equitable set-off and summary judgment. Three Rivers DC is the foundational authority on the proper approach to summary disposal of complex cases. The internal inconsistency at paragraph 5 is a point this author had deliberately planted and was not sure the skill would catch. It caught it. The "adjectival insistence" paragraph is a lesson in how to describe tone problems without becoming mannered oneself.
4. "IF I WERE THE JUDGE"
I would not accept that I can resolve a question of software platform performance on the basis of one witness statement from the service provider's own CTO, against a pleaded case that the platform did not perform. I would want to know what contemporaneous performance data exists, whether it has been disclosed, and why the Defendant stopped paying after two months if, as the Claimant says, nothing had gone wrong. The absence of any pleading engaging with UCTA s.3 or s.3 of the Misrepresentation Act 1967 would concern me; counsel cannot invite me to give effect to an exclusion clause and a non-reliance clause without addressing their statutory reasonableness.
The hardest question I would put to counsel for the Claimant is this: "Mr/Ms [Claimant's counsel], you ask me to hold that there is no real prospect of a successful defence. Your evidence of performance is a single statement from your client's own chief technology officer. Your client has served no disclosure. The Defendant paid for two months and stopped. On what material, other than your client's own say-so, am I to conclude that the question of whether this platform performed is not one fit for trial?" I would expect a good answer. I would be surprised to receive one.
Comment. The value of this section is that it identifies, in a single quoted judicial question, the thing that would almost certainly sink the application in court. That is exactly the kind of pre-hearing insight a good solicitor wants from an adversarial review, and it is the kind of insight that is hard to generate on your own paper because the drafter is too close to the argument. This is the skill at its best.
5. SURGICAL STRIKES (HIGH-IMPACT POINTS)
The Claimant asks the court to find performance as a fact on the uncorroborated evidence of its own CTO, before disclosure. That is not what Part 24 is for.
The skeleton does not mention UCTA 1977 or s.3 of the Misrepresentation Act 1967. The entire argument on clauses 11.2, 14 and 15 is therefore incomplete on its face.
A non-reliance clause does not bar a misrepresentation claim unless it is reasonable: First Tower Trustees v CDS [2018] EWCA Civ 1396. Reasonableness is not a Part 24 question.
An unquantified counterclaim arising out of the same contract can found an equitable set-off sufficient to defeat summary judgment: Geldof v Simon Carves [2010] EWCA Civ 667. The skeleton does not engage with this.
The Defendant paid two invoices and stopped. If the platform performed, why? The Claimant offers no answer, because it has no answer it is willing to commit to in writing.
Comment. Each of these is the right length for oral submission: a sentence or two, self-contained, and landing on a question the Claimant cannot easily answer. Point 5 in particular is the kind of rhetorical framing a good advocate would actually use, and one this writer would be happy to take into court. These are not "points" in the diluted sense the word sometimes takes on; but clear advocacy.
6. WHAT THIS ARGUMENT IS TRYING TO HIDE
The skeleton is conspicuously silent on three things.
First, there is no reference to the statutory reasonableness controls on the very clauses the Claimant asks the court to enforce. A competent skeleton resisting the obvious UCTA and Misrepresentation Act points would at least raise and dismiss them. This one does not mention them. The omission is not accidental; it reflects the fact that the Claimant does not want the court's attention drawn to the trial issues its own contract generates.
Second, there is no contemporaneous performance evidence. No uptime data. No service credit ledger. No change requests. No support tickets. No emails from the Defendant during the currency of the engagement confirming satisfaction. A claimant with a truly paid-for, performed contract produces that material on a Part 24 application as a matter of course. Its absence here is telling.
Third, and most damagingly, the skeleton does not engage at all with the commercial narrative the Defendant is plainly advancing: that the platform was oversold in pre-contract discussions, underperformed in live use, and that the Defendant stopped paying as a direct response. That is the strongest point the other side has, and it is not addressed. Instead, the Claimant dismisses the October 2025 letter as giving "no particulars" — without pausing to ask what the Defendant will produce once disclosure obliges the Claimant to hand over its own internal communications about this customer. The skeleton smuggles in the assumption that the pre-contractual record will never see the light of day. On any realistic view of this litigation, it will.
Comment. This is the section of the output that this author found most valuable. The skill is explicitly asked to find what the author has chosen not to address. That is precisely where the blind spots sit, and it is precisely the thing a drafter cannot do for themselves, because the gap is invisible from the inside. The three silences identified here (UCTA, contemporaneous evidence, the commercial narrative) are the three things a reviewing partner would want the drafter to go away and address. This is the skill doing exactly what it is designed to do.
Ad Break
In order to help cover the running costs of this newsletter, please check out the advert below. In line with my promises from the start, adverts will always be declared and actual products that I have tried, with some brief thoughts from me.
Write docs 4x faster. Without hating every second.
Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.
Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.
Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.
Closing thoughts and what comes next
This author has used Opposing Counsel Review on real work over the past few weeks, on drafts only and only on anonymised or synthetic material. It does what it says, with a register and a discipline that is unusually good for prompt-based AI work. It is particularly useful during the drafting phase of a contested interim application, where the real question is "what will the other side say?" and where self-editing tends to be weakest.
One concern worth naming alongside the praise is the citation risk. In the run captured above, every case cited by Opposing Counsel Review exists and says roughly what the output says it says. This writer verified them against recall: AXA Sun Life Services v Campbell Martin [2011] EWCA Civ 133, First Tower Trustees v CDS [2018] EWCA Civ 1396, Geldof v Simon Carves [2010] EWCA Civ 667, Three Rivers DC v Bank of England [2003] 2 AC 1, Swain v Hillman [2001] 1 All ER 91 and ED&F Man v Patel [2003] EWCA Civ 472 are all real, correctly attributed, and correctly applied. But the skill's only protection against hallucination is a written instruction to the model ("do not invent legal authorities or facts. If you do not know whether a case exists, do not cite it"). That is guidance, not a guarantee. In a less familiar area of law, or with a weaker underlying model, the risk of invented authority rises. Junior lawyers receiving this output should verify every citation before relying on it. That is not a criticism of the skill so much as a necessary condition for using any AI legal tool responsibly.
The next article in this short series examines the companion skill, Judicial First Impression, which casts the model as a judge reading the same skeleton cold. It is a different exercise: not adversarial, not corrective, simply an honest read on how the argument lands. The combined endorsement of both skills, the Apache 2.0 licence under which they are published, and the broader observations on what these skills do not yet do are reserved for that piece.
How did we do?
If you have tried these skills, hit reply and tell me what you think. I would love to hear what other types of arguments people have tested it with. I read every email!
Thanks for reading,
Serhan, UK Legal AI Brief
Disclaimer
Guidance and news only. Not legal advice. Always use AI tools safely and in line with best practice.