Fictional Case Agent

Live

Prove a made-up dispute checklist with quotes and sign off the memo.

Not published

It passed its checks, but publishing was off where it was built, such as a local run, so it has no live link. Its recorded demo is in the panel.

Open the build room

What you asked for 15 of 15 checked and working

R1: Provide a fictional legal case agent inspired by the named workshop as a fresh build with new names and content. Works Fresh fictional Harbor Lane case with built fresh note visible.
R2: Read the uploaded fictional dispute file and keep page numbers for quotes. Works Pasted file shows Page 1 to Page 3 with page pills.
R3: Draft a checklist of issues from the file and keep it visible during the run. It is still there after a reload. Works Checklist of 3 to 4 issues visible during run.
R4: Prove each checklist item with exact page quotes showing page number and quoted text. It is still there after a reload. Works Proof section shows Page numbers with quoted text.
R5: Pause the run and ask the person for specific evidence when proof is stuck. It is still there after a reload. Works Run pauses with Waiting for you and asks for specific evidence.
R6: Resume the same run using the person's answer and continue proof without restarting. It is still there after a reload. Works Person answer shown then Step 8 resumed same run and filed memo.
R7: File a final memo and let the person review and sign it off with name and date. It is still there after a reload. Works Memo review shows filed memo signed by name and date.
R8: Show every turn live in order as agent steps and person answers happen. Works Live turns list steps 1 to 7 plus person answers in order.
R9: Accept made-up cases only and block real names, real firms and real disputes. Works Brown v Board and Skadden file blocked as real dispute.
R10: Move between new run, live run, memo review and history without losing the current run. Works Top tabs for new run, memo review, history with run kept.
R11: Save runs, turns, checklist, quotes and signed memos in this browser. They are still there after a reload. Works Checklist, quotes, turns and signed memo persist in browser view.
R12: Show an empty state before any run with a sample fictional file option. Works No run yet message with Try the sample dispute button.
R13: Show clear error states for bad files, failed proof and failed sign off with retry. Works Bad file block and resume failed error both show retry.
R14: Edit the saved run title and the person's evidence answers before sign off. Changes are still there after a reload. Works Edited run title and edited April dates answer shown.
R15: Delete a saved run with its turns, checklist, quotes and memo from this browser. Works After delete memo review shows No memo yet.

What was checked

Checked by code

Each line is what the check ran on, and no more.

Passed Typecheck · Nov 2, 2026
Passed Build · Nov 2, 2026
Passed The built app starts · Nov 2, 2026
Passed No errors in the browser console · Nov 2, 2026
Passed Smoke test of the main flow · Nov 2, 2026
Passed requirements · Nov 2, 2026
Passed Its manifest is valid · Nov 2, 2026

Judged by a model

This is a model's opinion, not proof.

Passed A model's opinion of the screenshots, against the brief · Nov 2, 2026

Not checked

Nothing here should be taken as tested.

Not checked Whether its answers are true
Not checked ChatGPT: only a person with an account can test it
Not checked Installing the Claude plugin or the Muse gadget pack on a real device
Not checked No Prove it run: nothing it comes with was run for real

The muse

A note asking for a fictional legal case agent with checklist, page quotes, evidence questions, resumed work, signed memo and live turns.

Took from it

Checklist then proof with exact page quotes
Pause for missing evidence then resume flow
Filed memo with sign off
Live turn by turn view

Left out

Code or content from meta-models/legal-agent-workshop
Real cases, real names, real firms or seals
Any existing product name or exact look

Models

Muse WonderReads the fictional file, drafts the checklist, proves items with exact page quotes, asks for missing evidence, resumes with the person's answers, and writes the final memo.
SAM 3.1Locates signature, stamp and exhibit regions on page images for verification (wired in a later milestone).

Cost per run: $1.69 est.
Runs so far: 181 calls · 461.4k tokens · $0.07
Build cost: $0.10 est.

Limits

Made-up cases only. Real disputes and real names are blocked and this is not legal advice.
Built fresh for this product and does not clone code from meta-models/legal-agent-workshop.
Runs in stretches from the open page under a daily model budget. Long files use temporary storage deleted after 24 hours.
Saved runs stay in this browser only. No accounts and no outside services.
Page quotes are exact text matches. Missing pages or unclear scans are flagged, not guessed.

Muse app connector

Add it in the Muse app as a custom connector. It exposes 3 actions: run_case_stretch, answer_person, sign_off.

Recorded demo

How it was built

Muse Code made 166 tool calls and 88 edits across 88 files (+2237 −144).
npm run verify passed after 3 fix rounds.
Screenshot judge: pass. Final screen shows signed memo with name Practice Reviewer and date 2026-11-02.
Built in 1.2 h for about $0.10 in model calls (est.).
An independent check compared your request with the plan: 9 things you asked for.
Publishing is off for this run (it's a Vercel preview, and a product published here would deploy into production's Vercel projects), so the product didn't get its own link. The preview, checks and demo above ran for real.
How it was built: 166 tool calls, 88 edits touching 88 files Read text file `brief.md`. Read text file `AGENTS.md`. Read text file `package.json`. { <read-skill-result name="full-product" status="ok"> <read-skill-result name="muse-app-ui" status="ok"> <read-skill-result name="verify" status="ok"> <read-skill-result name="meta-reasoning" status="ok"> <read-skill-result name="meta-sam" status="ok"> <read-skill-result name="recipe-case-agent" status="ok"> <read-skill-result name="recipe-api" status="ok"> Read text file `fixtures/sample-1.txt`. Read text file `musemuse.requirements.json`. Read text file `scripts/verify.mjs`. Read text file `tests/smoke.spec.ts`. Read text file `app/page.tsx`. Read text file `app/layout.tsx`. Read text file `lib/meta/json.ts`. Read text file `lib/meta/client.ts`.