Jmail World - Search Epstein Files, E-Mails & Messages

Search and analyze the Jeffrey Epstein email archive (1.78M emails, 4,500+ iMessages, 18K photos, 1.4M documents, 473 identified people) via jmail.world's of...

Security Scan

The skill's code, requirements, and instructions align with its stated purpose (searching and querying jmail.world Parquet/APIs); it performs network fetches and local DuckDB queries as expected and requests no credentials.

Purpose & Capability

The name/description match the actual behavior: scripts query https://jmail.world API and https://data.jmail.world/v1/ Parquet files and run DuckDB. No unrelated credentials, binaries, or config paths are requested.

Instruction Scope

Runtime instructions and included scripts make HTTP requests to jmail.world and data.jmail.world, cache Parquet files under /tmp/jmail-cache, and run DuckDB queries. Inputs are sanitized and validated in the scripts, but user-supplied text is interpolated into SQL strings (mitigated by a whitelist sanitizer). The scripts can download large datasets and photos (sensitive personal data) — this is expected for the stated purpose but has privacy/storage implications.

Install Mechanism

No install spec; the skill is instruction-only plus included shell scripts. It checks for a local duckdb binary and instructs the user to install it manually. No remote code downloads or installers beyond fetching public Parquet/NDJSON files from data.jmail.world.

Credentials

The skill requests no environment variables, no credentials, and no config paths. Network access to jmail.world/data.jmail.world is required and proportionate to the stated data-retrieval purpose.

Persistence & Privilege

always is false and the skill does not request permanent platform-level privileges. It writes cached files to /tmp/jmail-cache (its own working cache) but does not attempt to modify other skills or global agent settings.

Assessment

This skill appears coherent with its purpose, but review and accept these points before installing: - Network & storage: it downloads Parquet/NDJSON and photo/document files from data.jmail.world and caches them under /tmp/jmail-cache. Expect potentially large downloads and disk usage. - Sensitive content & legality: the datasets include personal communications and images; ensure you have a lawful and ethical reason to access and store this material. - Safety of scripts: the included shell scripts are readable and perform input sanitization and validation, but they do interpolate sanitized values into SQL strings—run them in an isolated environment (container or VM) if you are unsure. - Authenticity: the SKILL.md references jmail.world; if you need higher assurance, verify the remote domain/URLs independently (open the site directly) and confirm the dataset origin before fetching large files. - Dependencies: the scripts require duckdb and curl (and optional jq/python3 for URL encoding). Install these from trusted package sources. If you are comfortable with these tradeoffs and the provenance of jmail.world, the skill is consistent with its claims; otherwise run it in a sandbox or decline installation.

jmail-world — Epstein Archive Search & Analysis

Search and analyze Jeffrey Epstein's email archive, iMessages, documents, photos, and people on jmail.world.

When to Use

• Researching Epstein connections and communications
• Finding emails between specific people
• Exploring iMessage conversations
• Analyzing communication networks and timelines
• Looking up people identified in photos
• Searching documents (DOJ releases, court records)

Data Access — Two Methods

1. Web Search API (quick lookups)

GET https://jmail.world/api/emails/search?q=QUERY&limit=50&page=1&source=all&from=NAME

No auth needed. Use web_fetch or curl. Good for quick keyword searches.

2. DuckDB + Parquet (deep analysis)

All datasets served as static Parquet files from https://data.jmail.world/v1/. No API key, no rate limit, no auth.

Use the included scripts:

# Full-text email search (Web API)
bash scripts/jmail-search.sh "query text"
bash scripts/jmail-search.sh "scopolamine" --from "Epstein" --limit 20

# DuckDB queries (local Parquet, no rate limits)
bash scripts/jmail-duckdb.sh search "flight manifest"
bash scripts/jmail-duckdb.sh person "Ghislaine Maxwell"
bash scripts/jmail-duckdb.sh imessages "ghislaine-maxwell"
bash scripts/jmail-duckdb.sh imessage-search "AfD"
bash scripts/jmail-duckdb.sh imessage-search "Merkel" --from "Bannon"
bash scripts/jmail-duckdb.sh imessage-list
bash scripts/jmail-duckdb.sh network "Bill Clinton"
bash scripts/jmail-duckdb.sh timeline "2005-01-01" "2005-12-31"
bash scripts/jmail-duckdb.sh top-senders
bash scripts/jmail-duckdb.sh people
bash scripts/jmail-duckdb.sh documents "flight"
bash scripts/jmail-duckdb.sh photos "person-name"
bash scripts/jmail-duckdb.sh photo-search "pool"
bash scripts/jmail-duckdb.sh photo-download "EFTA00000002-0.png" ./output
bash scripts/jmail-duckdb.sh download "EFTA02406146"
bash scripts/jmail-duckdb.sh download "HOUSE_OVERSIGHT_034601"
bash scripts/jmail-duckdb.sh stars

Available Datasets

Dataset	URL	Size	Records
Emails (full)	emails.parquet	334MB	1.78M
Emails (slim)	emails-slim.parquet	41MB	1.78M
Documents	documents.parquet	25MB	1.41M
Photos	photos.parquet	~1MB	18K
People	people.parquet	<100KB	473
Photo Faces	photo_faces.parquet	<100KB	975
iMessage Conversations	imessage_conversations.parquet	—	—
iMessage Messages	imessage_messages.parquet	—	—
Star Counts	star_counts.parquet	~2MB	414K
Release Batches	release_batches.parquet	<10KB	—

All under https://data.jmail.world/v1/.

Key Schemas

Email Columns (slim)

id, doc_id, sender, subject, to_recipients (json), cc_recipients (json), bcc_recipients (json), sent_at (timestamp), account_email, email_drop_id, epstein_is_sender (bool)

Email Additional (full)

content_markdown, content_html, attachments (int)

iMessage Conversations

id, slug, name, bio, photo, last_message, last_message_time, pinned, confirmed, source_files (json), message_count

iMessage Messages

id, conversation_slug, message_index, text, sender ("me" = Epstein, "them" = contact), time, timestamp, source_file, sender_name

People

id, name, source, photo_count

Documents

id, source, release_batch, original_filename, page_count, size, document_description, has_thumbnail

Web Pages (browser needed)

• /person/SLUG — Person profile
• /flights — Flight records
• /photos — Photo browser
• /drive/new-only — New documents
• /topic/SLUG — Topic pages

Document Full-Text Search

Documents have sharded full-text files (large downloads):

• documents-full/VOL00008.parquet — DOJ Volume 8
• documents-full/VOL00009.parquet — DOJ Volume 9
• documents-full/VOL00010.parquet — DOJ Volume 10
• documents-full/DataSet11.parquet — DOJ Dataset 11
• documents-full/other.parquet — House Oversight, court records

Query with DuckDB:

SELECT id, original_filename, extracted_text
FROM read_parquet('https://data.jmail.world/v1/documents-full/other.parquet')
WHERE extracted_text ILIKE '%rothschild%'
LIMIT 10;

Photo Columns (full schema)

id, source, release_batch, original_filename, content_type (MIME), width (px), height (px), image_description (AI-generated)

Search photos by description:

SELECT original_filename, image_description, width, height
FROM read_parquet('https://data.jmail.world/v1/photos.parquet')
WHERE image_description ILIKE '%pool%'
LIMIT 20;

Downloading Files

Universal Download (any document, email, or photo)

# Download by document ID — auto-detects type and source
bash scripts/jmail-duckdb.sh download "EFTA02406146"                    # DOJ email PDF
bash scripts/jmail-duckdb.sh download "HOUSE_OVERSIGHT_034601"          # House Oversight photo
bash scripts/jmail-duckdb.sh download "COURT_giuffre-115cv07433_1"      # Court document
bash scripts/jmail-duckdb.sh download "vol00009-efta00462570-pdf"       # DOJ volume scan
bash scripts/jmail-duckdb.sh download "EFTA02406146" ./output-dir      # Custom output dir

Supported sources:

ID Pattern	Type	Format
EFTA*	DOJ emails & photos	PDF
vol*	DOJ volume scans	PDF
HOUSE_OVERSIGHT_*	House Oversight photos	JPG (direct)
COURT_giuffre*	Giuffre court docs	PDF

Photo Search & Download

# Search photos by AI-generated description
bash scripts/jmail-duckdb.sh photo-search "swimming pool"
bash scripts/jmail-duckdb.sh photo-search "forced entry"

# Download a specific photo (DOJ: extracts PNG from PDF, HO: direct JPG)
bash scripts/jmail-duckdb.sh photo-download "EFTA00000002-0.png"
bash scripts/jmail-duckdb.sh photo-download "HOUSE_OVERSIGHT_034601.JPG" ./output

DOJ photos are stored as single-page PDFs — the download command automatically extracts the embedded image as PNG using pdfimages. House Oversight photos are direct JPGs. If pdfimages is not installed, DOJ photos are saved as PDF.

Security

• SQL injection prevention: All user input is whitelist-sanitized (alphanumeric + safe chars only). SQL meta-characters, operators, and keywords are stripped before query construction.
• Path traversal prevention: Parquet filenames and document IDs are validated against strict patterns before use in file operations.
• Read-only queries: All DuckDB operations are SELECT on read_parquet() — no writes, no code execution, no network access beyond the cached Parquet files.
• Data source: All data comes from jmail.world's public Parquet files and assets.getkino.com (DOJ document mirror). No private APIs or credentials involved.

Requirements

• DuckDB (duckdb CLI) — must be installed manually (apt install duckdb / brew install duckdb / duckdb.org)
• curl — for web search API and downloading Parquet files
• python3 — for URL encoding in search script
• pdfimages (optional, for photo-download PNG extraction) — apt install poppler-utils
• jq (optional, for JSON formatting)

References

• Official Docs
• Docs Index for LLMs
• API Reference
• DuckDB Examples
• Python Client (external) — not bundled, see official docs
• Datasets & URLs