docket-tracker
A Claude Code plugin that scaffolds and maintains a file-driven litigation-tracker website for any U.S. federal case, fed by CourtListener.
It is the generalized, reproducible form of the Anthropic PBC v. U.S. Department of War tracker. You point it at a case (a name or some CourtListener docket URLs), and it produces a Next.js 15 + MDX site that tracks every docket in the matter — district, circuit, and any parallel petitions — with importance-classified filings, RECAP PDF-availability badges, a timeline, parties, and a drafted legal explainer. No database: every fact lives in data/*.yaml or MDX and renders at build time.
A daily GitHub Action pulls new filings from CourtListener and commits them, and whatever host you connect (Vercel, Netlify, Cloudflare Pages, a self-hosted Node server, …) redeploys on push.
The honest 3-layer split
This plugin does not pretend a website about a lawsuit can be fully automated. It splits the work into three layers and is explicit about which is which:
| Layer | What it is | Who's responsible |
|---|
| 1. Data | Docket entries, importance labels, RECAP PDF status, parties, case metadata | Automated from CourtListener — pulled, classified, and committed daily with no human in the loop. |
| 2. News | RSS feeds + relevance terms per case; a digest of candidate stories | Configured by you. The engine is generic; you set the feeds and terms, and a human flips each item to approved: true before it shows on the site. |
| 3. Legal narrative | The case-explainer MDX, claims / issues / holdings / glossary, and the lib/citations.ts authorities | Drafted for a lawyer to review. Generated from primary sources, clearly marked as a draft, and never authoritative. |
⚠️ Disclaimer
The legal-narrative layer is a machine-generated draft, not legal advice and not a statement of law. It is produced from primary court documents to give a qualified attorney a head start, and it must be reviewed and verified by a licensed lawyer before publication. Every fact is meant to trace to a source; the plugin is instructed never to fabricate a holding, citation, quote, or docket entry. Treat anything in this layer as unverified until a competent professional has signed off. CourtListener mirrors PACER via RECAP, so very recent filings can lag — confirm against the official docket for anything time-sensitive.
Install
This repository is its own Claude Code marketplace. From inside Claude Code:
/plugin marketplace add willimj3/vaill-docket-tracker
/plugin install docket-tracker@docket-tracker
To hack on it locally instead, load it straight from a clone without installing:
claude --plugin-dir /path/to/vaill-docket-tracker
Restart Claude Code (or reload plugins) when prompted, and the two skills, the agent, and the CourtListener connector become available.
What you get
Skills
/docket-tracker:new \"<case name or CourtListener docket URLs>\" — Scaffolds a new tracker. Resolves the case to its set of dockets, copies the bundled template/ Next.js app into a target directory, writes data/case-meta.yaml and the per-docket entry files, populates the data layer from CourtListener, drafts the legal-narrative layer for your review, and runs npm run build as the gate. Leaves git init, committing, and deploying to you./docket-tracker:sync — Refreshes an existing tracker. Pulls new CourtListener entries, classifies their importance, appends them (labeled high / medium / low, nothing dropped) to data/dockets/*-entries.yaml, and refreshes the RECAP-status badges. This is the same logic the daily GitHub Action runs — use the skill for an on-demand pull or to backfill.
Agent
narrative-drafter — A subagent the new skill delegates to for the legal-narrative layer. It reads the complaint and key opinions from CourtListener and drafts the case explainer, claims, issues, holdings, and glossary — always marked as a draft for the lawyer's verification, with every fact traced to a primary source.
Bundled template — template/ is the complete Next.js 15 + MDX app the new skill copies. File-driven, no database, deployable as-is to any Next.js host (Vercel, Netlify, Cloudflare Pages, a self-hosted Node server, or a static export to any static host).
The headline: adding a docket is one edit
The docket set is driven entirely by the dockets: list in data/case-meta.yaml. Each entry is just (ids and courts below are illustrative — use your own case's values):
dockets:
- id: district
court: D. Example
level: trial # trial | appellate
courtlistener_id: 10000001
- id: appeal
court: 9th Cir.
level: appellate
courtlistener_id: 10000002
