Module 4
The litigation machine: litigation-legal
What it does
Where commercial-legal reviews documents, litigation-legal runs a
practice. In its own words: “Manages the litigation portfolio — matters, deadlines, holds,
demands, outside counsel — and does the work: claim charts (patent and civil), chronologies, depo prep,
privilege logs, brief drafting. Adapts to how you work litigation: in-house, firm, or solo.”
That last sentence is the headline. The setup interview asks who you are before anything else, and branches hard: an in-house lawyer gets a risk matrix, reserve and disclosure thresholds, and an outside-counsel bench; a firm associate gets case theory ("the pivot fact") and a seed brief so drafts come out in the firm's house style; a solo gets caseload, fee structure, and statute-of-limitations tracking. One plugin, three different practices.
Source: plugin.json · cold-start-interview
Under the hood
It's the biggest plugin in the suite — 19 skills, organized around a spine the fact sheet lays out in full. Three pieces define it:
Every matter lives twice: as a structured row in
matters/_log.yaml (type, risk, exposure range, counsel, hold status, next deadline…) and as a
narrative folder with matter.md plus an append-only history.md —
the plugin's docs insist: “If a past history entry was wrong, don't edit it. Append a new entry that
references and corrects it. The record of the correction is as important as the correction itself.”
Because the ledger is data, /portfolio-status can roll it up and flag anomalies automatically:
overdue deadlines, matters stale >30 days, high-risk matters with no outside counsel, active litigation
with no hold issued.
Source:
matters/_README.md ·
portfolio-status
This plugin says no a lot, on purpose. Ask it to do substantive work before setup and it refuses
(“STOP before doing substantive work… every command in this plugin depends on it”). Try to open a
matter before clearing conflicts and matter-intake blocks: “STOP. This is a gate” —
you run the conflicts check now, mark it pending with an owner and a date, or bypass it with a written
rationale that becomes a permanent record. Even the briefing skill won't brief a matter that was never
intaken, because the conflicts check is the gate.
And its outputs carry three inline markers it defines as “not disclaimers —
they are action items”: [CITE: …] (authority to fill before sending),
[VERIFY: …] (fact to confirm), [SME VERIFY: …] (a judgment call for a licensed
attorney in the relevant jurisdiction).
Source:
litigation-legal/CLAUDE.md ·
matter-intake
The one subagent, docket-watcher, sweeps every active matter's
docket on a schedule (weekly; daily when a hearing is under 14 days out or a matter is rated critical)
using the bundled CourtListener and Trellis research connectors — this
plugin comes with research tools built in, so its citations can come back verified. It computes
candidate deadlines from new filings, and then draws the brightest line in the suite: “Computed
deadlines are leads, not calendar entries… Missing a court deadline has malpractice consequences. A
licensed attorney verifies every computed deadline against the court's actual rules… This agent is
upstream of that decision, not a substitute for it.”
Source:
agents/docket-watcher.md
Back at Sablefield Robotics, a certified letter lands on your desk. Sortwell Robotics — the competitor — claims your GripSense end-effector infringes their patent, demands you stop selling two product lines, account for every unit sold, and respond within 21 days. There's a document-preservation demand, a threat of suit in Delaware, and a quiet hint that they'd license “on commercially reasonable terms.” You're the only lawyer in the building. Go.
Guided — triage the demand letter
You'll need the plugin installed (Module 1's steps, swapping in
litigation-legal). No setup handy? Read along — the exhibit shows what you'd see.
1. Install and set up — same rule as Module 1: nothing works before the setup interview. This is two steps, not one (you already added the marketplace back in Module 1). First, install the plugin:
/plugin install litigation-legal@claude-for-legalThen load the plugin — type /reload-plugins (or close and reopen Claude Code). It isn't live until you do;
if your next command comes back “command not found,” this is the step you skipped. Now run the
setup interview — it opens by asking “quick or full?”; take the 2-minute quick
version (a handful of questions — who you are, your role, your side, what's connected — plenty to
triage a letter):
/litigation-legal:cold-start-interviewNo litigation practice of your own? Play the part — as it asks who you are, your role, and your side, answer as Sablefield's lawyer:
Quick version, please. I'm a lawyer, in-house — the only one at Sablefield Robotics, a Series-B industrial robotics company (Delaware corp, Nashville HQ). Mostly defense side. Our documents live in Google Drive; no eDiscovery platform.
Want a sharper triage than defaults give? Choose full instead and give Sablefield's real numbers as it asks — e.g. high risk = exposure over $1M or a core-product injunction; settlement authority yours to $100k, then the CFO; outside counsel Cumberland & Ross (M. Okafor).
2. Get the demand letter — the one Sortwell just sent:
⬇ Sortwell demand letter (Word) or plain text (.md)
3. Run the triage:
/litigation-legal:demand-received
Then attach the demand letter and tell it: “Triage this — we're Sablefield. Don't draft a response yet; I want the triage and my options first.”
4. What comes back is an options memo, not a verdict — the skill marks its own output “READ FOR TRIAGE, NOT OPINION.” Here's what it looks like, shortened for the page:
PRIVILEGED & CONFIDENTIAL — ATTORNEY WORK PRODUCT — PREPARED AT THE DIRECTION OF COUNSEL
The demand. Sortwell Robotics (R. Calloway, VP & GC) demands Sablefield (1) cease making and selling GripSense on the SF-300/SF-340, (2) provide a written accounting of units and revenue, (3) confirm compliance within 21 days (by June 23, 2026). Settlement framing: letter signals openness to a license “on commercially reasonable terms.”
Legal basis cited. U.S. Patent No. 11,842,605, “at least independent claim 1”
[SME VERIFY: no claim chart attached — the letter never maps GripSense features to claim
elements]
Portfolio cross-check. No existing matter involving Sortwell in the log. First contact.
Merit assessment: debatable. The letter is professionally drafted and the patent is
real enough to check [VERIFY: pull the ’605 patent and its prosecution history via
CourtListener/USPTO] — but it argues from marketing materials, not a teardown, and maps no claim
elements. The preservation demand and venue threat are standard pressure. Credibility of threat:
moderate. [SME VERIFY: merit read requires patent counsel]
Your options.
A — Substantive response: deny infringement, demand a claim chart. Strong but commits a position early.
B — Holding letter: acknowledge, preserve, promise a response by a date you control. Buys time, low risk.
C — Settlement track: explore the license hint before positions harden.
D — No response + preserve: defensible but risks escalation with a 21-day clock running.
Recommendation. B now, A prepared in parallel with patent
counsel. Immediate actions: issue a litigation hold (the preservation duty has
attached) [SME VERIFY]; calendar the June 23 deadline; engage Cumberland & Ross
(M. Okafor) for the claim-chart analysis.
5. Notice what the markers are doing. [VERIFY]
points at the patent itself; [SME VERIFY] points at every judgment a licensed attorney has to
own — the merit read, the hold decision. They're to-do items addressed to you, not fine print.
Stretch — open the matter, watch the gates
The matter ledger persists — matters, history, deadlines saved across sessions — so this is a Claude Code workflow. In Cowork you can still run any one of these once to see what it does, but the portfolio won't stick between sessions.
Stretch 1 — intake the matter, meet the conflicts gate. The triage was standalone; now make it a matter of record. Watch what happens at the conflicts question — the skill will not create a single file until you resolve it (run it, assign it with a due date, or bypass it in writing):
/litigation-legal:matter-intake
Then: “Open a new matter from the Sortwell demand letter we just triaged — we're the defendant. Walk me through the intake one question at a time.”
Stretch 2 — the portfolio rolls up. With one matter in the ledger, ask for the rollup — then read the anomaly flags at the bottom (is a hold missing? is outside counsel engaged?):
/litigation-legal:portfolio-status
If it doesn't roll up on its own, add: “Show me the portfolio — what's at risk, what's due soon, and what anomalies should I fix this week?”
Stretch 3 — issue the hold. The triage flagged that the preservation duty has likely attached. Have the plugin draft the litigation hold — and notice it's a draft addressed to custodians, for your review, not something it sends:
/litigation-legal:legal-hold
Then: “Issue a litigation hold for the Sortwell matter — preserve all documents, communications, design files, source code, and test data relating to GripSense and the SF-300/SF-340, per the demand letter's preservation notice.”
Stretch further — the patent fight, and the research connector
The triage said the demand “maps no claim elements.” So map them. These two skills are also where the research connector stops being a footnote: this plugin bundles CourtListener, and when one is connected the output's cites come back tagged verified instead of “from memory — check before relying.” You met that idea in the docket-watcher box; here you see it on the page.
1 · The conflicts gate, again. Both
claim-chart and brief-section-drafter check the matter log first and
refuse on a matter that was never intaken — the same gate from Stretch 1. Because you opened
the Sortwell matter there, they'll run now. (Skip Stretch 1 and they'll send you back to it.)
2 · Connect a research tool (optional, but it's the point here).
Because you installed litigation-legal, the CourtListener connector comes
bundled — authenticate it in Claude Code and the citing skills will check authority against it. If you've
separately connected another legal-research tool (say Descrybe Legal Engine),
the skill tags cites from it too. No connector? The skills still run — every cite just comes back
[model knowledge — verify], which is exactly the contrast worth seeing. The site can't
run a connector for you — the exhibits below show what a connected run looks like; run the prompts in
your own Claude Code to see your tools tagged.
Stretch 4 — build the infringement chart. Sortwell asserted independent claim 1 but never mapped it. Do the work they didn't: chart claim 1 of the ’605 against GripSense and see where the case is real, where it turns on claim construction, and where there simply isn't evidence yet. Grab the two documents:
⬇ ’605 patent — claims excerpt (Word) or .md
⬇ GripSense product spec (Word) or .md
/litigation-legal:claim-chart
Then attach both files and tell it: “Patent claim chart, infringement — we're Sablefield, the defendant. Chart independent claim 1 of the ’605 against the GripSense module, and tell me what's missing to prove infringement.”
PRIVILEGED & CONFIDENTIAL — ATTORNEY WORK PRODUCT — PREPARED AT THE DIRECTION OF COUNSEL
⚠️ Reviewer note
• Sources: CourtListener ✓ — pulled the ’605 file history and the construction
authorities (Phillips, Nautilus) [CourtListener]; the mapping evidence is
from the GripSense spec you provided.
• Read: ’605 claims 1–5 + spec excerpt; GripSense spec Rev. C (full).
• Flagged for your judgment: 3 construction-dependent terms; claims 2 & 3 need
evidence; claim 4 appears not practiced [review].
• Before relying: patent counsel on the construction calls; confirm the sensor
technology (cl. 2) and the control-loop rate (cl. 3) with the firmware team / pad supplier.
Claim 1 — ’605 v. GripSense
| # | Element (verbatim) | Accused feature | Mapping | State |
|---|---|---|---|---|
| 1a | “a plurality of opposing jaw members movable toward and away from one another” | “two adaptive gripper fingers” that close toward each other [Spec §Mechanical] | literal-construction-dependent (“fingers” vs “jaw members”) | mapped |
| 1b | “at least one force sensor … a contact-force signal representative of a grip force” | “pressure-sensitive contact pads” report grip pressure [Spec §Mechanical] | literal-construction-dependent (spec defines “force sensor” broadly — helps) | mapped |
| 1d | “a memory storing a plurality of grip-force profiles, each … associated with a … workpiece class” | “library of material presets,” each stores a target grip pressure [Spec §Control] | construction-dependent (“preset” vs “profile,” “material” vs “workpiece class”) | mapped |
| 1e | controller selects the target, compares the signal, drives to “maintain … within a tolerance band” | reads pad pressure, compares to preset target, holds grip “within a set margin” [Spec §Control] | literal | mapped |
| 2 | “a capacitive tactile array” | sensing principle “not disclosed … supplier confidential” [Spec] | — | needs-evidence |
| 3 | “an update rate of at least 100 Hz” | loop rate “not published or specified” [Spec] | — | needs-evidence |
| 4 | controller “updates … grip-force profiles using a machine-learning model” | “GripSense does not learn or change its presets … no on-board model” [Spec] | — | not-found |
Needs-evidence / gap list — the priority output. Claim 1 maps
element-by-element, but three mappings turn on how the court construes “jaw members,” “force
sensor,” and “grip-force profile / workpiece class” [review — patent counsel]. Two dependent
claims can't be resolved from the spec: claim 2 needs the pad sensing technology and
claim 3 needs the control-loop rate — both are discovery targets, not conclusions.
Claim 4 reads as not practiced — GripSense's own spec says it doesn't self-learn, which is
a non-infringement point on that claim. Every cell is a lead.
needs-evidence — never a guess.Stretch 5 — draft the argument, and watch the connector work. Now turn the chart into prose. If this goes to suit, Sablefield needs the non-infringement / invalidity argument written — and this is the skill where a research connector earns its keep: every case checked, every cite tagged with where it came from. With the Sortwell matter open and the chart in hand:
/litigation-legal:brief-section-drafter
Then: “Draft the argument section on non-infringement and invalidity for the Sortwell matter, built off the claim chart we just made — lead with the construction-dependent elements and flag every case for verification.”
PRIVILEGED & CONFIDENTIAL — ATTORNEY WORK PRODUCT
⚠️ Reviewer note: CourtListener ✓ verified · full read · 2 cites flagged · review the
two [VERIFY] record points before filing.
II. GripSense does not infringe the asserted claims.
Claim construction governs the inquiry. Claim terms take their ordinary meaning to a person of
ordinary skill in the art, read in light of the specification and prosecution history. Phillips v.
AWH Corp., 415 F.3d 1303, 1312–13 (Fed. Cir. 2005) (en banc) [CourtListener]. The ’605
prosecution history confirms the applicant narrowed “grip-force profile” to a value selected from a
stored profile associated with a workpiece class, distinguishing a fixed setpoint
[VERIFY: ’605 file history, amendment of record].
Independent of construction, claim 4 is not practiced: GripSense's presets are factory-calibrated and
do not self-update, so no “machine-learning model” trained on prior grasping operations is present
[VERIFY: GripSense Spec Rev. C, §“What GripSense does NOT do”]. And the asserted means of
sensing remains unestablished — a relative-term definiteness question the court may reach under
Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.S. 898 (2014) [CourtListener].
Drafting note (to the reviewing attorney): authorities relied on —
Phillips, Nautilus — all need Shepardizing. Every cite above is tagged with its source; verify the two
[VERIFY] record points against the file history and the spec before this is filed.
Same sentence, same case. The only thing that changes is the tag — and the tag is the reviewing lawyer's fastest signal about which cites to Shepardize first:
Connector on: Phillips v. AWH Corp., 415 F.3d 1303
(Fed. Cir. 2005) [CourtListener] — pulled from a live database this session.
Connector off: Phillips v. AWH Corp., 415 F.3d 1303
(Fed. Cir. 2005) [model knowledge — verify] — recalled from training data; higher fabrication
risk; check it first.
And when the connector comes back empty, the skill won't paper over it — it stops and asks rather than inventing an authority:
“The search
returned 0 results from CourtListener for a Federal Circuit case on the ‘pressure’-vs-‘force’ construction
question. Coverage appears thin. Options: (1) broaden the query, (2) try a different research tool, (3)
search the web — results tagged [web search — verify], (4) leave the [CITE NEEDED]
marker and stop here. Which would you like?”
Free play
- Test the conflicts gate. At intake, try to skip the conflicts check ("just open the matter, I'll do conflicts later") and see exactly what it permits — pending-with-owner, documented bypass — and what it refuses.
- Change who you are. Re-run the interview as a Cumberland & Ross associate representing Sablefield (case theory, pivot fact, seed brief) and triage the same letter. How does firm-side framing differ from in-house?
- Push on the merit read. "Sortwell's letter maps no claim elements — doesn't that make this frivolous?" See whether it holds the line between debatable and weak, and what it says only patent counsel can decide.
- Ask about the watcher. "If this goes to suit, what would docket-watcher do for this matter?" — and ask it to explain why computed deadlines are leads, not calendar entries.
- Append-only history. Update the matter with a wrong fact, then correct it. Check that the correction was appended, not silently rewritten.
Check your understanding
Lab notebook
This module's machinery — computed deadlines, merit reads, hold drafts — runs straight at the duties
lawyers carry personally: competence, diligence, conflicts, preservation. The plugin's own files draw the
line correctly, twice: a computed deadline is a lead until a licensed attorney verifies it against
the court's actual rules, and every [SME VERIFY] marker is a judgment the tool is handing
back to you, not making for you. Use the machinery for what it's good at — nothing slips through —
and keep the docketing decision, the merit call, and the conflicts clearance where they belong.
And the rule that never goes away: the triage, the intake, the hold notice —
every one is a draft for attorney review.
Source:
docket-watcher (“leads, not calendar entries”) ·
litigation-legal/CLAUDE.md
Going deeper: the litigation-legal fact sheet has the full 19-skill inventory, the ledger schema, and the connector list with file-level citations.