Your first chronology from your own matter, then a prompt for every skill.
Demo data. A fictional matter.
The three steps from the video. The full detail for each is further down.
In Cowork, click Customize in the left sidebar and open the plugins list (Anthropic's README calls it Browse plugins). Search legal, pick Litigation Legal and add it with the button shown, Add or Install.
Paste the command below. It asks Quick or Full first: Quick takes about 2 minutes, Full about 10 to 15. Then it asks your role (in-house, at a firm, or solo) and the side you mostly represent (plaintiff, defense, both with a default, or varies by matter). For a litigator at a midsize firm, answer at a firm and pick Midsize firm as the practice setting.
/litigation-legal:cold-start-interview
Start a new conversation, choose the folder with your documents, and paste this.
/litigation-legal:chronology --documents The production is the folder I have open, [first to last Bates number]. Date each email in a thread by its own message, not the file name, and merge an event that appears in several documents into 1 row citing every Bates number. Put undated documents under Gaps instead of giving them a date. Make the output a visual artifact that makes it very easy for me to understand the key events in order, the gaps, and which Bates number backs each date.
Click a step for the why and the tips.
Why. The plugin installs inside Cowork, so Cowork has to be on your account first.
Anthropic's Quick Start in the claude-for-legal repo links an install video for Cowork.
Why. Installing gives you the 19 litigation skills as slash commands and the docket watcher agent.
Type / in the chat box and check that the /litigation-legal: commands appear. If they do not, close and reopen the app; in Claude Code the documented fix for Command not found is a restart. If you are ever asked to pick a scope, choose all projects (user scope), because project scope blocks the plugin from reading files outside the project folder.
/litigation-legal:cold-start-interview
Why. Every skill stops until this interview has written your practice profile. Role decides which questions run: the firm path asks for your case theory, the pivot fact, a seed brief and your document review setup; the in-house path asks about reserves and board memos instead.
Quick writes working defaults marked [DEFAULT] that you can tune later. Full adds your real risk bands, outside counsel roster, house brief style, privilege log format and demand letter templates. Say pause at any point and it saves your answers so you can resume.
Why. It learns your citation format, headings and tone from the brief and only asks about gaps, so every later draft reads like your firm wrote it.
With fewer than 10 seed documents it marks the profile LIMITED DATA: skills still run, with thinner output. If you state a rule, deadline or threshold during setup, it checks it before writing it into the profile.
Why. Every skill reads this file before it does anything, and any skill stops while a [PLACEHOLDER] marker remains.
It is plain English and survives plugin updates. The CLAUDE.md inside the plugin folder is only a template that gets replaced on every update. Company facts go to company-profile.md one folder up, shared with the other legal plugins.
/litigation-legal:cold-start-interview --check-integrations
Why. Without a research tool every citation comes from training data, is tagged verify, and the reviewer note says sources were not verified. With it, a citation carries a [CourtListener] tag only when CourtListener actually returned it in that conversation.
The check marks a connector connected only after a real tool call succeeds; configured but untested shows as not verified. CourtListener covers U.S. court opinions and PACER dockets, not statute text. The demand drafter still does not look up authority on its own, so ask for it in plain words.
Why. The chronology reads paths you give it in the session first, and the plugin reads only files you point it at or that sit in the current folder.
Keep anything that should not be read, such as your own notes, out of that folder. If a skill says it cannot read a file, the usual causes are project scope, a typo in the path, or a format it cannot open.
/litigation-legal:chronology --documents
Why. Documents mode works from a production with no matter slug, so it runs before any matter intake. Matter mode refuses a matter that has not been through /litigation-legal:matter-intake, because intake is where the conflicts check is recorded.
Option A is for sources already privilege screened; with B every row carries ok, flag or review. It closes by asking you to scan the red entries for anything it miscalled. Check each key date against its Bates number before you rely on it.
/litigation-legal:customize
Why. You adjust one position without re-running the whole interview or editing the file by hand.
It will not remove the privilege and FRE 408 gate on demand letters, the privilege header or the source tags. For a full re-interview use /litigation-legal:cold-start-interview --redo.
Click a skill to see what to have ready and the prompts. Paste one into Claude and fill in the brackets.
/litigation-legal:chronologyHave ready. A finished practice profile and the documents: a Bates-numbered production, custodial files or an eDiscovery export (--documents), or a matter already run through matter intake (--matter). PDFs, .eml emails, .docx and .txt are read directly; Everlaw or Aurora through a connector.
Modes.
--documents Production mode, the default for a firm associate. Extracts from the documents with Bates citations and needs no matter slug.--matter Matter mode, the default in-house. Reads the matter's theory and declared document sources, and refuses a matter not in the matter log.--format=sof Statement of facts variant: key and relevant events as prose with record cites. Privilege-flagged entries are left out by default.--format=witness-[name] Witness variant: only events where that person sent, received, attended or is the subject.--include-flagged Keeps privilege-flagged entries in the statement of facts variant, recorded in the header as a permanent acknowledgment.Production chronology, as a visual artifact
/litigation-legal:chronology --documents The production is the folder I have open, [first to last Bates number]. Date each email in a thread by its own message, not the file name, and merge an event that appears in several documents into 1 row citing every Bates number. Put undated documents under Gaps instead of giving them a date. Make the output a visual artifact that makes it very easy for me to understand the key events in order, the gaps, and which Bates number backs each date.
Threads and duplicates are where a chronology goes wrong, the skill's own rules are to merge 1 event with several sources and never invent a date, and the artifact puts the pinpoint for each row where you can check it.
Statement of facts skeleton, flagged rows left out
/litigation-legal:chronology [matter-slug] --format=sof Use the key and relevant entries only, in date order, 1 paragraph per event with its Bates number or page:line. Keep the privilege-flagged entries out and list what you left out, so I can decide on each one.
The statement of facts variant drops flagged and review entries by default until counsel confirms privilege, which makes it the safe skeleton for a brief's fact section.
/litigation-legal:claim-chartHave ready. A matter already intaken, the side, the forum and the phase. Civil: the cause of action or defense and the pleading. Patent: the patent number and at least 1 asserted claim.
Modes.
--civil Civil element chart for any cause of action or affirmative defense, with a gap list framed for the phase.--patent Patent claim chart, element by element against a target.--infringement Patent sub-mode: claim elements against an accused product.--invalidity Patent sub-mode: claim elements against prior art.--review Audits a chart or brief someone else produced, element by element.provisional Say this if your profile is not set up yet: it runs on generic US defaults and tags every row [PROVISIONAL].Breach of contract, gaps as discovery priorities
/litigation-legal:claim-chart --civil Matter: [matter-slug]. Count: breach of contract. Side: [asserting / defending]. Forum: [court or arbitral seat], governing law [state]. Phase: discovery. [attach the pleading, the agreement, and the key documents or deposition excerpts] Confirm the element list with me before mapping, cite every cell by Bates number or page:line, and turn the gap list into discovery priorities.
Elements vary by jurisdiction and the skill confirms them before mapping; in discovery each gap becomes a deposition, a document request or an interrogatory.
Audit the other side's chart
/litigation-legal:claim-chart --civil --review [attach opposing counsel's brief or chart] For each element, check whether their cited evidence actually supports it, mark partial support, and list where their chart is thinnest.
Review mode checks someone else's work element by element, and a cite that backs only part of an element is the error it is built to catch.
/litigation-legal:demand-intakeHave ready. A finished practice profile and the facts of the demand. No document is required, but the contract, invoice and prior correspondence make the draft sharper.
Modes.
--full Runs the strategic block (leverage and BATNA, downside, settlement communication posture, privilege filters, admission risk) whatever the materiality heuristic says.Payment demand, posture answered up front
/litigation-legal:demand-intake [client] unpaid invoice --full Posture: measured tone, [14]-day response window, no settlement marking, signed by [attorney name]. Demand type: payment. Recipient: [the other side], through opposing counsel [name, firm]. Triggering event: invoice [number] dated [date], $[amount], net [30], unpaid. [attach the invoice and the agreement] Ask me the rest, and name the thin spots before you save.
The skill asks tone, window, marking and signer first and will not fall back to a default for any of them, so answering in the first message takes you straight to the core questions.
Privilege filters and the admission trap
Before you save the intake, walk the strategic questions on 3 things with me: whether this letter is a settlement communication under FRE 408 or the state equivalent in [forum], what from our internal analysis must stay out of the letter, and anything the other side could later call an admission or an accord and satisfaction, such as a partial payment marked paid in full.
The skill exists for the pre-writing a careless letter skips, and these are the 3 places a letter waives privilege or concedes something.
/litigation-legal:demand-draftHave ready. A saved intake from demand-intake (it refuses without one, or with an empty strategic block on a material demand) and your answer to each of the 7 gate items.
Modes.
--skip-gate Bypasses the pre-draft checklist when it was run and documented separately. The draft file records that it was skipped and why.--version=N Writes draft-vN.docx instead of the next version number.Draft from the intake, gate answered item by item
/litigation-legal:demand-draft [slug] Run the pre-draft gate 1 item at a time and wait for my answer on each. In the draft, cite only the contract sections and documents in the intake, mark every fact not yet tied to a document [VERIFY], and leave every legal authority as a [CITE] placeholder.
The skill will not draft on blank answers, and it treats a misquoted clause or an invented cite as the fastest way to lose credibility with opposing counsel.
Fill the citations from CourtListener, tagged
Fill the [CITE] for [the proposition, for example the payment obligation under the governing state's law] from CourtListener and tag each one with where it came from. If the results are thin, stop and give me the options instead of filling it from memory. Leave any statute cite as a placeholder.
The skill does not research citations on its own, CourtListener returns court opinions rather than statute text, and a thin search comes back as options rather than a guess.
/litigation-legal:demand-receivedHave ready. The incoming letter as a file. A matter log makes the cross-check useful.
Modes.
--slug=custom-slug Names the inbound folder instead of the generated slug.Triage with our own decision date
/litigation-legal:demand-received [path to the letter] Note whether they marked it as a settlement communication and whether the substance is actually a compromise discussion. Flag every authority they cite for me to verify rather than accepting it, and set our internal decision date ahead of their stated deadline.
A label alone does not make a letter a settlement communication, the skill never validates the other side's law itself, and their deadline does not bind you but your decision date should.
Hand off the option I pick
We will go with option [A / B / C / D]. Hand it off with the fields pre-filled to [matter intake / demand intake for our response], and if there is no legal hold on this dispute, flag that first.
Each option routes to its next skill with the triage carried forward, and even setting a demand aside still means preserving documents.
/litigation-legal:deposition-prepHave ready. A matter already run through matter intake (it refuses otherwise), the case theory and pivot fact in your profile, and the witness name. Its document pull runs through a connected eDiscovery platform such as Everlaw; without one, give it the documents yourself.
Adverse witness, the topics that matter
/litigation-legal:deposition-prep [witness name] Matter: [matter-slug]. Posture: adverse, so closed leading questions, 1 fact each. Their documents: [Bates range, or attached]. Keep it to the 3 or 4 topics that matter most, lead with the strongest confrontation, and cite every document by Bates number. Put any prior statement in as a paraphrase marked [verify exact quote] unless you have the passage and its page:line in front of you.
The skill sets question form by witness posture, calls a sprawling outline unfocused, and never puts quotation marks around words it cannot pin to the record.
Impeachment list with split cites
For [witness name] in [matter-slug], list every document where they wrote or received something that conflicts with what we expect them to say, with the Bates number and the specific conflict. Split any cite that supports only part of the point.
An impeachment falls apart when the pinpoint supports only part of the proposition, which is the failure the skill's record fidelity rule is written against.
/litigation-legal:privilege-log-reviewHave ready. A matter already intaken, the log file or document set, and your log format and review protocol from setup. It first asks whether the documents came through disclosure or discovery.
Check the log before production
/litigation-legal:privilege-log-review [attach the log] Matter: [matter-slug], forum [court or arbitral forum]. Check that every entry has date, author, all recipients, document type, privilege claimed and a usable description. Keep the designation on anything close and give me the specific question behind each flag.
Missing fields are flagged before the substantive pass, and the skill never strips a designation on its own read.
A single call that clears many flags
Group the flagged entries by the question they turn on, for example the same third party copied or the same in-house lawyer giving business and legal advice, and tell me how many entries each decision resolves.
The pattern pass is where a single attorney call can resolve 50 flags, so review time goes where it counts.
/litigation-legal:matter-intakeHave ready. A finished practice profile, the matter basics and the conflicts status. The initiating document (demand, complaint, demand for arbitration, subpoena) sharpens it.
Open the matter with conflicts on the record
/litigation-legal:matter-intake [matter name] Counterparty: [the other side]. Our role: [claimant / respondent / plaintiff / defendant]. Forum: [court or arbitral institution, case number]. Source: [how it arrived, with the date]. [attach the initiating document] Conflicts: cleared by [name] on [date], checked against [the other side, its affiliates, opposing counsel, key witnesses]. Show me the log row and matter.md before you write anything.
The skill will not create a matter while conflicts are not run, and deposition prep, briefs, holds and claim charts all refuse a matter that is not in the log.
Theory as a working hypothesis
For the initial posture, our story: [2 sentences]. Theirs: [2 sentences]. The pivot fact: [1 sentence]. Mark the theory as a working hypothesis, and if no legal hold has been issued, flag it so I can run the hold next.
The skill records the theory you give it without inventing one, marks it for attorney verification, and flags a missing hold on active or anticipated litigation.
/litigation-legal:matter-briefingHave ready. A matter in the log with its matter.md and history.md. It reads; it does not write.
Before the call
/litigation-legal:matter-briefing [matter-slug] I have a call with [the supervising partner / the client's general counsel] in [30] minutes. Tailor the last section to the questions I should ask, the decisions I need and the updates to get, and flag anything stale.
Naming the purpose turns on the section for the conversation, and a matter with no update in 30 days is flagged at the top.
Is the risk rating still right
/litigation-legal:matter-briefing [matter-slug] Focus on the risk re-assessment check: what in the recent history suggests the risk or materiality field no longer fits the matter. Give it to me as questions, not a new rating.
The skill surfaces the question and leaves the call to counsel; changing the rating is a matter update, not a briefing.
/litigation-legal:matter-updateHave ready. A matter in the log and the event: type, date and a short summary.
Log a new deadline
/litigation-legal:matter-update [matter-slug] [the tribunal set the expert disclosure deadline for date] Event type: procedural. Update the next deadline and the stage, and show me the history entry and the log change before writing.
Procedural events usually touch only the stage and the next deadline, and every field change is shown to you before it is written.
Settlement demand received
/litigation-legal:matter-update [matter-slug] [written settlement demand received from opposing counsel on date] Event type: strategy. Record it as received, not accepted, and ask me the materiality question explicitly.
Settlement activity forces an explicit materiality answer, and logging an offer does not trigger the gate that accepting one does.
/litigation-legal:matter-closeHave ready. A matter in the log, the resolution, and 2 or 3 honest sentences of lessons.
Close after a settlement
/litigation-legal:matter-close [matter-slug] Resolution: settled on [date], $[amount], [structural terms]. Compare the final cost to the exposure range at intake. Lessons: [your 2 or 3 sentences]. Remind me if the legal hold still needs a release.
The outcome is the data point that calibrates future intakes, and closing a matter does not release its hold, which runs separately.
Consolidated into another matter
/litigation-legal:matter-close [matter-slug] Resolution: consolidated into [parent-slug] on [date]. Leave the lessons empty unless I give you some.
Consolidation records the parent matter, and the skill leaves lessons blank rather than inventing them.
/litigation-legal:matter-workspaceHave ready. Matter workspaces turned on (a private-practice setting at setup), a lowercase hyphenated slug, the client, the other side, the matter type, a confidentiality level and the key facts.
Modes.
new <slug> Runs a short intake and writes matter.md, history.md and notes.md. It does not switch to the matter on its own.list Table of matters with status, marking the active one.switch <slug> Makes that matter active, so skills read its facts and write to its folder.close <slug> Archives the matter. It is never deleted.none Detaches from any matter and works at practice level.Open a workspace with its overrides
/litigation-legal:matter-workspace new [client-matter-2026] Client: [client]. Counterparty: [the other side]. Matter type: [contract dispute]. Confidentiality: heightened. Key facts: [2 to 5 sentences]. Override for this matter only: [for example measured tone, the other side is a long-standing supplier].
A matter override beats the practice default, and cross-matter reading stays off so this client's facts stay in this folder.
Switch before you start work
/litigation-legal:matter-workspace list Then switch to [matter-slug] and show me its summary so I can confirm it is the right matter.
The active matter decides which folder every skill reads and writes, so switching first keeps the work filed in the right place.
/litigation-legal:brief-section-drafterHave ready. A matter already intaken, the case theory and a seed brief captured at setup, and the inputs for the section: the chronology and key documents for facts, the issue and authorities for argument.
Statement of facts from the chronology
/litigation-legal:brief-section-drafter statement of facts Matter: [matter-slug]. Written submission to [the panel / the court]. Build it from the statement of facts chronology and cite every fact to a Bates number, page:line or exhibit. Facts only, no argument, and mark anything unconfirmed [VERIFY].
The skill asks written or oral first, and its statement of facts rule is selection and sequence with a record cite on every fact, never argument.
Argument section, weak points named
/litigation-legal:brief-section-drafter argument [II] Issue: [the issue]. Authorities we rely on: [list, or leave CITE NEEDED]. Address the other side's best counterargument, flag any point where the authority cuts against us instead of dressing it up, and tag every citation with where it came from.
The skill is built to say when an argument is weak and to tag each cite by source, so the reviewing attorney knows which cites to check first.
/litigation-legal:subpoena-triageHave ready. The subpoena as a file. A grand jury subpoena stops the skill and escalates to criminal counsel.
Modes.
--slug=custom-slug Names the inbound folder instead of the generated slug.Non-party document subpoena, deadlines first
/litigation-legal:subpoena-triage [path to the subpoena] Our client is not a party. Research the rule for [forum] first and show me the objection deadline counted from service and from the compliance date, with the rule cite tagged by source, before the scope analysis.
Objection deadlines often run from the earlier of 2 dates, and the skill researches the forum's rule first instead of assuming a number.
Objections framework, strength rated
For each category requested, give me the objections that apply with their basis, the custodians they cover and a strong, reasonable or weak rating, then what we would likely produce after objections. Flag whether a protective order is needed for confidential material, separately from privilege.
The skill produces the framework, not the objections letter, and it treats confidentiality and privilege as separate grounds.
/litigation-legal:legal-holdHave ready. A matter already intaken (except for --status), plus scope, custodians, date range and systems. Your hold template from setup if you have one.
Modes.
--issue First notice: captures scope, custodians, date range and systems, drafts the .docx and sets the next refresh (default 6 months).--refresh Reaffirmation notice with scope and custodian changes. Flags departed custodians as an IT preservation action.--release Release notice with the retention instruction, once the matter is truly over.--status Report across every matter: active holds, overdue refreshes, active matters without a hold, closed matters with a hold still on.Issue a hold with the scope spelled out
/litigation-legal:legal-hold [matter-slug] --issue Scope: [subject matter] from [start date] to present. Custodians: [names and roles]. Systems: [email, shared drives, messaging, phones]. Raise anything you think the scope misses as a question, and show me the notice and the log change before writing.
The skill proposes scope and leaves the call to you, because too broad is a burden and too narrow is spoliation risk.
Every hold at a glance
/litigation-legal:legal-hold --status Put first any active matter with no hold issued and any refresh that is overdue, then the closed matters where a hold is still on.
The status report surfaces 3 of the failures the skill is built against: a hold never issued, never refreshed, or never released.
/litigation-legal:portfolio-statusHave ready. A matter log with matters in it, built by matter intake.
Modes.
--all Includes closed matters.--risk=high Filters by risk band (critical, high, medium or low).--stale Only matters with no update in more than 30 days.--type=[type] Filters by matter type, for example employment.--owner=[name] Filters by business, HR or comms owner.Monday read, anomalies first, as a dashboard
/litigation-legal:portfolio-status Lead with the anomalies: overdue deadlines, conflicts pending or bypassed, active or threatened matters with no hold. Then build the dashboard view: the summary line on top and 1 sortable table.
The anomaly rules are what make the rollup useful, and the plugin offers a dashboard for data-heavy output instead of building one unprompted.
Only what has gone quiet
/litigation-legal:portfolio-status --stale For each matter, show the last update date, the next deadline and the outside counsel lead, so I can pick which to update first.
Stale matters are where deadlines slip, and the skill surfaces them for you to prioritize rather than deciding for you.
/litigation-legal:oc-statusHave ready. A matter log with the outside counsel firm and lead partner email on each matter. Gmail connected only if you want Gmail drafts.
Modes.
--all Drafts for every active matter regardless of recency.--slug=[slug] Drafts for 1 matter only.--no-gmail Markdown drafts only, even when Gmail is connected.A single matter, markdown only
/litigation-legal:oc-status --slug=[matter-slug] --no-gmail Ask about [the open item from matter.md, for example the draft reply brief] and where the budget stands against the authorization. First name and bullets.
The draft follows your house tone and only asks about what the matter file actually holds, and it waits for you to send it.
Weekly sweep, run summary first
/litigation-legal:oc-status Show me the run summary first: which matters got a draft, which were skipped and why, and any anomalies such as a missing lead email.
The summary names every skip, so a matter never drops out of the weekly check without you seeing why.
/litigation-legal:cold-start-interviewHave ready. About 2 minutes for Quick or 10 to 15 for Full, plus what you have of: a prior brief in house style, the case theory memo, a privilege log format, a hold template, exemplar demand letters.
Modes.
--full Upgrades a quick start to the full interview.--redo Re-runs the interview on a configured profile. Add a section name to redo 1 part.--check-integrations Re-tests your connectors and refreshes the integrations table without re-interviewing.--new-matter Reuses the practice profile on a new matter (firm associate or solo).Full setup as a midsize firm litigator
/litigation-legal:cold-start-interview Full, not quick. I'm a lawyer at a midsize firm doing briefs, discovery and deposition prep, mostly on the [defense / plaintiff] side. I'll share a prior brief in our house style and our privilege log format when you ask for documents.
Giving role, practice setting and side up front routes the interview to the firm path: case theory, pivot fact, seed brief and review setup.
Re-check connectors after adding CourtListener
/litigation-legal:cold-start-interview --check-integrations I just connected CourtListener.
It marks a connector connected only after a real tool call succeeds, and it updates the profile without a new interview.
/litigation-legal:customizeHave ready. A finished practice profile and the one change you want.
Change the default side
/litigation-legal:customize Change my default side from [both, default plaintiff] to [defense]. Tell me what changes in demand letters, chronologies and claim charts before you write it.
It shows the current value and the downstream effect, then confirms, and it flags tensions the change creates elsewhere in the profile.
Update the house brief style
/litigation-legal:customize Update our house style: [citation format, heading convention, tone]. [attach a recent brief]. Flag anything in the profile that now conflicts.
House style feeds every brief section draft, and one change at a time avoids repeating the whole interview.
Every output is a draft for attorney review: not legal advice, not a legal conclusion, not a substitute for a lawyer.
Source: Anthropic's claude-for-legal repo, README.md (repo root)
Every skill stops until the setup interview has written your practice profile. Only the interview itself and --check-integrations run without it.
Source: Anthropic's claude-for-legal repo, litigation-legal/CLAUDE.md
The chronology will not extract until you choose a privilege posture, and it never invents events or fills thin coverage from model knowledge without asking.
Source: Anthropic's claude-for-legal repo, litigation-legal/skills/chronology/SKILL.md
The demand drafter does not send the letter and does not research citations on its own: an authority you did not supply stays a [CITE] placeholder.
Source: Anthropic's claude-for-legal repo, litigation-legal/skills/demand-draft/SKILL.md
A citation carries a research tool's tag only if that tool returned it in the conversation. Everything else is tagged model knowledge, verify.
Source: Anthropic's claude-for-legal repo, litigation-legal/CLAUDE.md
The privilege log review never strips a designation on its own read. Close calls keep the designation and go to the attorney.
Source: Anthropic's claude-for-legal repo, litigation-legal/skills/privilege-log-review/SKILL.md
You can follow these steps yourself. If you hit a wall, book 30 minutes with us.