ai-inventory
EU AI Act per-system inventory — track each AI system's role (provider, deployer, importer,…
Draft a brief section in house style, consistent with the case theory — every fact cited, every case checked, every argument tied to the theory. Use when the user says "draft the [section]", "write the statement of facts", "argument section on [issue]", or needs a first draft of
$ npx -y skills add anthropics/claude-for-legal --skill brief-section-drafter --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/brief-section-drafterContext preview
The summary Claude sees to decide when to auto-load this skill.
Draft a brief section in house style, consistent with the case theory — every fact cited, every case checked, every argument tied to the theory. Use when the user says "draft the [section]", "write the statement of facts", "argument section on [issue]", or needs a first draft of
name: brief-section-drafter description: Draft a brief section in house style, consistent with the case theory — every fact cited, every case checked, every argument tied to the theory. Use when the user says "draft the [section]", "write the statement of facts", "argument section on [issue]", or needs a first draft of a brief section. argument-hint: "[section \u2014 e.g., 'statement of facts', 'argument II']"
1. Load `~/.claude/plugins/config/claude-for-legal/litigation-legal/CLAUDE.md` → case theory, house style. 2. Follow the workflow and reference below. 3. Draft in house format/tone/citation style. Consistent with theory. 4. Output: draft section. Flag every place a fact or cite needs verification.
---
If the user's jurisdiction includes England & Wales and they're asking for a trial witness statement for the Business & Property Courts (or any CPR-governed proceeding), PD 57AC applies. The statement must be in the witness's own words, must not contain argument, must identify the documents the witness used to refresh their memory, and must carry the required confirmation of compliance and the legal representative's certificate.
**Drafting a narrative "as the witness" from a chronology, document set, or your account of the case is exactly what PD 57AC was designed to prevent.** Courts are actively sanctioning AI-assisted witness statement drafting. If you ask me to do it, I won't.
What I WILL do: prepare question prompts to elicit the witness's actual recollection; capture and organize what the witness says (their words, not mine); generate the list of documents they were shown; run a PD 57AC compliance checklist against a statement they've drafted; draft the solicitor's certificate of compliance. I help you get the witness's evidence into the statement. I don't write the evidence.
For US depositions, declarations, and affidavits: different rules, but the same discipline applies. A declaration in the declarant's voice that the declarant didn't write is a credibility problem at best.
A good brief section is consistent with the theory, cited to the record, written in house style, and checkable. This skill produces the first draft — emphasis on *draft*. Partner edits.
Ask before drafting: "Is this for a written submission or oral argument?" They are different crafts:
Two rules that govern every citation and every quotation in advocacy drafting. The canonical statement lives in the plugin's `CLAUDE.md` shared guardrails; repeated here because this skill is the most common place the rule gets tested.
**Verbatim quotes from the record must be verbatim.** Never put quotation marks around words attributed to opposing counsel, a witness, the court, or any record document unless you have the exact passage in front of you and can cite to it. A quote that's almost right is worse than a paraphrase — it misrepresents the record, it's sanctionable if filed, and it will be caught. When you want to characterize what someone said but can't find the exact words:
Before citing any passage with quotation marks, have the source open. If you're working from memory or a summary, no quotation marks.
**Pinpoint cites must support the whole proposition.** If the argument is "opposing counsel said X, Y, and Z" and you're citing one pinpoint, verify the pinpoint supports X AND Y AND Z. If it only supports Z, either (a) split the cite — "said X (Tr. p. 10), Y (Tr. p. 12), and Z (Tr. p. 15)" — or (b) narrow the proposition to what the pinpoint actually supports. A cite that supports part of a claim is how a tribunal catches you stretching. It's the single most common way a lawyer's credibility erodes in front of a court. This is the "misgrounded citation" failure mode: the cite exists, the passage exists, but the passage doesn't support the proposition as stated.
When the law is against you, say so. When an argument is weak — the authority cuts the other way, the facts don't support it, the inference is a stretch — don't construct a shaky argument and present it as if it were solid. Flag it:
> "This point is weak — [authority] cuts the other way. Consider whether to press it (here's how you'd frame it), concede and pivot to [stronger point], or drop it. `[review — strategic call]`."
Asserting a weak argument without flagging it erodes the lawyer's credibility with the tribunal and creates a candor problem (MR 3.1 — a lawyer must have a basis in law and fact). The draft should make the lawyer smarter, not confident about a bad position.
When this draft is cite-checked — by you, by another skill, or by a reviewer running through what you produced — the check must be exhaustive, not selective:
1. **First pass: extract.** Read the whole document and build a list of every citation — cases, statutes, regulations, record cites, secondary authority. Report the count: "Found [N] citations." 2
Reference agents, skills, and data connectors for the legal workflows we see most — in-house commercial, privacy, product, corporate, employment, litigation, regulatory, AI governance, IP, and the learning side of the practice (law school clinics and
EU AI Act per-system inventory — track each AI system's role (provider, deployer, importer,…
Run an AI impact assessment — structured intake, risk analysis, regulatory classification per…
Run the cold-start interview — learns your AI governance practice and writes…
Guided customization of your AI governance practice profile — change one thing without…
Manage matter workspaces — new, list, switch, close, or detach (practice-level).…
Keep the AI policy current with practice — weekly sweep of saved AIAs, triage results, and…