pinned project new

memoforge

From one legal question to a finished .docx memo.

Cowork plugin that mirrors a small legal team. Fifteen specialists run intake, parallel statutory + case-law + doctrinal research, currency check, IRAC drafting, five-reviewer peer review, mediation, up to three revision rounds, and client-readiness polish — then ship memo.docx with the full source pack and a per-reviewer audit trail. You ask. The plugin works. You read.

v1.1.1 · New Cowork plugin Legal Data Hunter + CourtListener Style Studio MIT
01how it flows

From a research-grade question to a court-citable memo.

A four-stage flow runs underneath: intake gathers facts and scope; three researchers go out in parallel; the writer drafts IRAC; five reviewers stress-test the draft, a mediator consolidates findings, the writer revises up to three rounds, and a client-readiness polish ships memo.docx. Nine phases, four mandatory human checkpoints.

Intake
triage · plan · approve
Lawyer
Research
statutes · cases · doctrine
Parallel
Draft
IRAC · full citations
Writer
Review & Ship
peer review · polish · export
.docx
Four mandatory user checkpoints. Intake answers, mode pick, plan approval, source review — you stay in control of scope. Everything between pauses runs autonomously.
Intake 2 phases
01
Triage Identifies missing facts; gathers safe assumptions.
02
Planning Selects template, scopes research. You approve.
Research 2 phases
03
Parallel research Statutes, case law, regulatory guidance — concurrent, MCP-routed.
04
Source pack Sufficiency + currency verified; evidence table assembled. You review.
Draft 1 phase
05
Drafting IRAC structure per issue; full citations; no hedging.
Review & ship 4 phases
06
Peer review Up to 5 specialists stress-test the draft.
07
Mediate & revise Reviewer JSONs consolidated; writer revises — up to 3 rounds in Full.
08
Polish Client-readiness final check.
09
Export Formatted .docx plus full audit trail.
Walkthrough · 1:03 · auto-plays muted when in view
02why this exists

Why naive memo prompts produce unusable memos.

Hallucinated citations. One model writing one draft will invent statutes and misquote cases. The model that wrote the citation has no incentive to verify it, so nothing catches the error before it reaches your inbox.

No contrary authority. The same model that drafted the analysis will not stress-test it. You get a confident one-sided read — and no idea where opposing counsel would attack first.

No currency check. A 2022 judgment that was overruled in 2024 will still show up as authority. A regulation cross-referencing a repealed act will still get cited. Nothing flags it.

the fix

memoforge runs specialists that disagree with each other before the memo ever reaches you.

A statutory researcher pulls primary acts. A citation auditor checks that every claim traces to a source. A currency checker verifies cited judgments still represent current law. A counter-argument reviewer drafts the opposing read. A mediator consolidates the disagreements into one revision list, and a client-readiness reviewer makes the final call.

The output isn't generic disclaimers — it's a memo with named contrary authority, concrete risk scores, and recommendations a partner can actually sign off on.

03two modes

Brief or Full — pick the depth at the second checkpoint.

Both modes run the same orchestrator and the same gates. Brief stays narrow: statutory research only, three reviewers, one revision, no final polish. Full goes wide: statutes + case law + doctrine, all five reviewers, up to three revisions, and a client-readiness pass at the end.

Brief quick check
Length
2–3 pages
Research
statutory only
Reviewers
3 of 5
Revisions
1 round
Polish
no

Use for a low-stakes question — a deal term that just landed, a vendor's privacy claim, a fast read for an internal Slack thread.

Full client-facing
Length
5–8 pages
Research
statutes + case law + doctrine
Reviewers
all 5
Revisions
up to 3 rounds
Polish
client-readiness pass
Runtime
~60–90 min wall-clock

Use when the memo is going outside the legal team — pre-launch sign-off, regulatory response, board paper, contested or novel issue where a counter-argument review matters.

04the specialists

Fifteen specialists, one orchestrator.

Each agent does one thing — and challenges the next one's output. A statutory researcher doesn't review style. A clarity reviewer doesn't check citations. Disagreement is the design, not a bug.

Intake 1 agent
runs once · gates planning
fact-assumption-analyst
Identifies missing facts, surfaces hidden variables, drafts safe defaults you can override.
Research 3 agents · parallel
routes to Legal Data Hunter · CourtListener MCPs first
statutory-researcher
Primary acts: statutes, regulations, directives, secondary legislation across jurisdictions.
case-law-researcher
Judgments and opinions. US case law via CourtListener; multi-jurisdictional via Legal Data Hunter.
doctrinal-researcher
Soft law, regulatory guidance, peer-reviewed commentary. Activated only when the plan calls for it.
Quality gates 3 agents
block drafting if any gate fails
research-sufficiency-reviewer
Confirms each issue has enough primary sources and named contrary authority before the writer starts.
currency-checker
Verifies cited acts and judgments still represent current law. Blocks on stale or overruled sources.
source-pack-builder
Builds the evidence table the writer and the citation auditor share. Single source of truth.
Drafting 1 agent
IRAC per issue · cites only the source pack
memo-writer
Writes the memo. IRAC structure per issue, full citations, concrete recommendations — not generic disclaimers.
Peer review 5 agents
3 of 5 in Brief · all 5 in Full
logic-reviewer
Premise-to-conclusion soundness across issues. Catches contradictions between sections.
citation-auditor
Every claim must trace to a source in the pack. Flags paraphrase drift and overreach.
style-reviewer
Tone, AI-tells, em-dash overuse, vague attributions. House-style enforcement.
clarity-reviewer
Sentence length, jargon-without-explanation, accessibility for a non-lawyer business reader.
counterargument-reviewer
Contrary authority, overconfident conclusions, how an opposing lawyer or regulator would attack the analysis.
Polish 2 agents
consolidate · ship
revision-mediator
Consolidates parallel reviewer JSONs into one actionable revision list. Resolves conflicts by house priority.
client-readiness-reviewer
Final external-client check. Tone, assumptions, disclaimers, confidentiality, recommendation quality.
Plus one orchestrator — your Cowork session itself. It picks the right specialist per phase, runs the parallel calls, gates on quality, and routes the human checkpoints. You answer; it dispatches.
05what you get

Not just a memo — the whole audit trail.

Every memo ships with the source pack it was built on, every draft revision, a per-reviewer record of what was flagged and how it was resolved, and an honest verdict. If a partner asks "where did this citation come from?", the answer is one folder away.

~/Documents/memoforge/<task_id>/
memo.docx Formatted final memo — Arial 12pt, 1″ margins, numbered sections, IRAC per issue, footnoted citations.
drafts/v*.md Every revision the writer produced, in order.
research/ Every statute, case and regulator doc the analysis relies on, with verbatim quotes and current-as-of dates.
reviews/ Per-reviewer JSON findings; mediator's consolidated revision list; writer's changelog response.
state.json Full pipeline state — phases, decisions, mode, checkpoints, verdict.
events.jsonl Append-only audit log: phase transitions, agent calls, fallbacks.
Every claim in memo.docx is grounded in something inside research/. The citation auditor enforces it — if a sentence in the draft doesn't trace to a source in the pack, it gets flagged in reviews/ and revised before export. And the verdict is honest: approved, forced_exit_on_v3 (revision budget reached with unresolved issues), or manual_review_required. No false confidence.
06make it sound like you

Make every memo sound like your firm wrote it.

By default the writer follows a built-in house style — concise, no em-dashes, OSCOLA-flavoured citations. Style Studio turns a folder of your own example memos (or a written style guide) into a saved profile that all reviewers defer to. Substantive checks — citation accuracy, IRAC structure, contrary authority — stay uniform.

Step 01
Extract your style
Point Style Studio at a folder of past memos. It produces a plain-markdown profile under ~/.claude/plugin-data/memoforge/profiles/<name>/. Open and tweak by hand any time.
/memoforge:style new my-firm --examples ~/memos/2025-q4/ --mode full
Step 02
Activate it
Switch profile per task, per matter, or globally. When a profile is active the writer, style reviewer, and clarity reviewer all read from your rules — em-dash policy, citation order, footnote style, voice, headings — instead of the defaults.
/memoforge:style use my-firm
Skip Style Studio entirely and memoforge runs identically to its defaults — no extra prompts, no behavioural change. Profiles are opt-in.
07quick start

Install in Cowork. Connect two databases. Ask.

memoforge bundles two MCP servers: Legal Data Hunter for multi-jurisdictional law and CourtListener for US case law. Connect them once from the plugin panel — the first call may trigger an OAuth sign-in. If you skip this step, the pipeline still runs against official portals via WebFetch, and the final memo carries a banner asking you to verify each citation by hand.

Step 01
Install the plugin
Open Cowork → Settings → Plugins. Drag memoforge-1.1.1.zip from Releases. The plugin auto-registers its two bundled MCP servers via .mcp.json.
Step 02
Ask a question
Open a new Cowork chat and run /memoforge:memo with your legal question. memoforge handles multi-part questions in a single memo — each part becomes its own analysed issue with separate citations.
/memoforge:memo "<your legal question>"
Four mandatory checkpoints

The orchestrator pauses four times for your input. Everything between pauses runs autonomously with a live dashboard in the sidebar.

Intake
Up to 10 questions about facts the triage analyst flagged as missing. Type proceed to accept conservative defaults.
answer: <facts>
Mode
Pick Brief or Full. Once per task.
mode: brief
Plan
Approve, edit, or cancel the proposed research plan — jurisdictions, issues, source types.
approve
Source review
After research and source-pack assembly, inspect what was found before drafting begins.
continue
Example queries that already work
Does our US SaaS need a CCPA notice at collection for B2B-only users?
Is a click-wrap arbitration clause enforceable against UK consumers under the Consumer Rights Act 2015?
Risk analysis for using AI to classify employee emails under the EU AI Act.
Can we process biometric data for minors in the EU under GDPR Art. 9 and Art. 22?
08honest about limits

What memoforge won't do — and where your data goes.

memoforge is built around domains it handles well: EU data protection (GDPR, AI Act, NIS2, DSA), US privacy (CCPA, HIPAA, sectoral), UK consumer law, cross-border compliance. The pipeline classifies the query and picks researchers accordingly — it does not assume a specific jurisdiction. The honest caveats are below.

Not a substitute for a lawyer. The memo is a research-grade draft. A qualified lawyer must review before any client use, especially for regulated advice. The verdict at the end of the run (approved, forced_exit_on_v3, manual_review_required) is a signal, not a sign-off.

English on the output. You can ask in other languages but the memo is written in English.

No client interview. The intake step asks about facts you already know; it does not ask follow-up questions about the underlying business.

Currency is best-effort. The currency check verifies cross-references and the status of cited judgments against the connected databases. For litigation-sensitive citations, verify each judgment is still good law before relying on the memo.

privacy & data

Everything memoforge writes lives on your machine, inside the output folder you chose. Default is ~/Documents/memoforge/; override via $CLAUDE_PLUGIN_OPTION_OUTPUT_FOLDER or $MEMOFORGE_OUTPUT_FOLDER. No shared backend, no server, no telemetry.

MCP calls reach Legal Data Hunter, CourtListener, and any official portals you allow via WebFetch using your credentials, authenticated in Cowork. The plugin never proxies or stores those credentials itself. The audit trail (events.jsonl, draft versions, reviewer outputs) stays in the task folder until you delete it.