agentleFS
Wang200935Wang200935/security-agent-skills/skills/recon/email-osint-investigation/SKILL.md

email-osint-investigation

Skill4 starsChanged 3 months agoMITView on GitHub

Email-first OSINT investigation — 7-track parallel workflow when the

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After all tracks complete, cluster findings by anchor attributes:

    line 136
  2. Very specific:

    Reddit .json 403: Use old.reddit.com/user/<name> instead.

    line 157
  3. Very specific:

    username-scanner --profile arg: Only quick/full/aggressive.

    line 163
  4. Distinctive:

    After running all 7 tracks, all free breach APIs have hit-count + field metadata + masked values.

    line 173
  5. Distinctive:

    If the user authorises the spend, use Playwright to drive the website (their account in their session) and harvest the unmasked data.

    line 178

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Email-First OSINT — 7-Track Parallel Investigation

When to Load This Skill

Load when the investigation seed is an email address (<user>@<domain>), not a username, domain, or phone number. Email-first OSINT has a fundamentally different shape than username-first SOCMINT because the email itself is a cross-platform anchor — it can be queried against service-registration databases (holehe), breach corpora (HudsonRock), and commit metadata (GitHub) that username scans cannot reach.

The 7-Track Workflow

Seed: <email>@<domain>
  ├── Track 1: holehe ────── email registered on which 120+ services?
  ├── Track 2: username-scanner ── local-part as username sweep across 840+ sites
  ├── Track 3: HudsonRock ── infostealer breach correlation (username + email) ← HIGHEST VALUE
  ├── Track 4: Gravatar ──── MD5(email) → profile JSON (avatar, linked accounts)
  ├── Track 5: GitHub API ── user profile + commit author-email search
  ├── Track 6: Direct platform checks (Instagram GraphQL, Pinterest JSON-in-HTML, Steam XML)
  └── Track 7: Multi-engine web ── DuckDuckGo HTML + Yahoo TW (NOT Google/Bing — captcha)

All tracks are independent → run in parallel. Batch into one execute_code script with urllib.request. Do NOT run serial terminal curl calls — the consent gate blocks some URLs unpredictably.

Track Details

Track 1: holehe (email service registration)

~/.hermes/hermes-agent/venv/bin/holehe <email> 2>&1 | head -100
  • [+] = registered (email confirmed present — strongest signal)
  • [-] = inconclusive
  • [x] = not registered

Pitfall: holehe runs ~500s. Always run in background=true with notify_on_complete=true.

Track 2: username-scanner (username sweep)

~/.hermes/hermes-agent/venv/bin/username_scanner <local-part> --no-nsfw --profile full --format json --output /tmp/osint_<name>/

Parse the nested JSON (variations.<user>.sites.<platform>.ai_analysis.probability), NOT a flat results list. Filter by probability >= 0.80, then manually verify each top-15 hit by direct HTTP GET.

Cross-variant check (same final_url returns Found for 2+ username variants) is the strongest real-account signal. See the username-scanner skill for the full filtering recipe.

Track 3: HudsonRock Infostealer (HIGHEST VALUE)

# By username (local-part of email):
curl -s "https://cavalier.hudsonrock.com/api/json/v2/osint-tools/search-by-username?username=<local-part>"

# By email (run BOTH — sometimes one returns data the other doesn't):
curl -s "https://cavalier.hudsonrock.com/api/json/v2/osint-tools/search-by-email?email=<email>"

Returns:

  • computer_name — often a person's first name (e.g. "Yun")
  • operating_system — e.g. "Windows 10 build 19045 (64 Bit)"
  • ip — partially masked (first 2 octets → ipinfo.io for city/country)
  • top_passwords — partially masked patterns (e.g. w********8)
  • top_logins — partially masked secondary emails/usernames on same machine — KEY PIVOT
  • date_compromised
  • total_user_services — count of stolen credentials

Why this is the highest-value track: The top_logins array often reveals a second email (e.g. education email 1*******@fsps.kh.edu.tw) that the primary email search would never find.

Taiwan school email deobfuscation: <id>@<code>.kh.edu.tw → resolve <code>.kh.edu.tw via https://www.<code>.kh.edu.tw/ → extract <title> for school name. Example: fsps.kh.edu.tw → 高雄市左營區福山國民小學.

Track 4: Gravatar

import hashlib
h = hashlib.md5(email.strip().lower().encode()).hexdigest()
# Profile JSON: https://www.gravatar.com/{h}.json
# Avatar:       https://www.gravatar.com/avatar/{h}

Most Gmail users don't have a Gravatar. A 404 here is normal and not a negative signal.

Track 5: GitHub

Two endpoints:

  1. https://api.github.com/users/<local-part> — user profile (404 if no account)
  2. https://api.github.com/search/commits?q=author-email:<email> — commits with this email (requires Accept: application/vnd.github.cloak-preview+json header)

Track 6: Direct Platform Checks

PlatformMethodReturns
InstagramGET /api/v1/users/web_profile_info/?username=X with x-ig-app-id: 936619743392459full_name, bio, followers, avatar, is_private, is_verified
PinterestGET /<username>/ → regex extract JSON from HTMLfull_name, country, locale, counts, created_at
RedditGET old.reddit.com/user/<name> (NOT /about.json — 403 to bot UA)404 = not found, 200 = exists
SteamGET /id/<name> HTML (NOT ?xml=1 — XML returns false negative)persona name, location, level, privacy
KaggleGET /<name>/reCAPTCHA → use Playwright
ReplitGET /@<name>HTML; holehe is more reliable

Track 7: Multi-Engine Web Search

EngineBehaviorUseful?
GoogleCaptcha block from non-CE IP❌ Useless with curl
BingCaptcha block / empty⚠️ Rarely
DuckDuckGo HTML"No results" = reliable negative✅ Best free signal
Yahoo (search.yahoo.com)Open, no captcha, CJK-friendly✅ Good for Taiwan names
Brave Search429 after 1-2 requests⚠️ Use sparingly

Do NOT mistake Google captcha page HTML containing the email in the search box <input value="..."> as a "hit" — it's just the query echoed back.

Cross-Platform Attribute Clustering

After all tracks complete, cluster findings by anchor attributes:

AnchorExampleWhat it proves
Email → HudsonRock computer_namewheat20131@gmail.com → "Yun"Person's name
Email → HudsonRock IP geo61.227.. → Kaohsiung, TWPhysical location
Email → HudsonRock secondary email1*******@fsps.kh.edu.tw → 福山國小School/employer
Pinterest countryTWConfirms country
Instagram full_name"wheat2013"Display name
Password patternw********8Credential risk

Confidence tiers:

  • Confirmed: ≥2 independent anchors match
  • Probable: 1 anchor + consistent metadata
  • Candidate: Username match only
  • Different person: Conflicting anchors

Pitfalls

  • Tavily 432: web_search and web_extract fail with HTTP 432. Use execute_code with urllib.request.
  • Google captcha: Returns captcha via curl. Email in <input value="..."> is NOT a hit — it's the query echoed back.
  • Reddit .json 403: Use old.reddit.com/user/<name> instead.
  • Steam XML false negative: ?xml=1 returns <error> for privacy-restricted profiles. Use HTML endpoint.
  • username-scanner inflated "Found": Filter by ai_analysis.probability >= 0.80 + manual verify. 128 raw Found → ~5-10 real after filtering.
  • terminal curl consent gate: Some curl commands trigger "BLOCKED". Use execute_code (Python urllib).
  • HudsonRock masking: Passwords/emails partially masked but first char + length preserved — often enough to pivot.
  • holehe [+] is verified: A [+] result is a confirmed registration, not a guess. Treat as confirmed in the report.
  • username-scanner --profile arg: Only quick/full/aggressive. Not basic/standard/intermediate.
  • DeHashed "Quick Check" free aggregate only: dehashed.com/search?query=<email> via Playwright (Cloudflare wait 15s → real page computes exposure count like "14 incidents") gives ONLY the count without login. Full records behind login.
  • LeakCheck /api/public returns structured hits without auth: {"success":true,"found":N,"fields":["password","id","origin"],"sources":[{"name":"Stealer Logs"}]}. Count + field names + source = free. No unmasked variant via any parameter (raw=true, full=true, key=public all return same masked shape).
  • LeakPeek free alias grid is the hidden gem: Despite PRO-tier pricing for raw row click, LeakPeek's "Alt Alias" grid partially unmasks values — patterns like Wheat*****, 7bca****, 10***@fs**.tw are directly visible. These partial patterns include prefix reveal that HudsonRock and LeakCheck fully mask. The 8-row alt-alias grid may show secondary emails/usernames/password-shapes the primary records obscure.
  • IntelX free search shows file names + redacted ██ data blocks but no unmasked records. "Full Data" buttons are PRO-gated (click does nothing free).
  • Ahmia + OnionLand via Tor work, but breach content is NOT free on the dark web either. Tor is principally for anonymous SE queries, not "free breach dump access".
  • Cloudflare-protected breach sites (DeHashed, breachdirectory.org, HIBP) can be passed via Playwright headless + 10-15s wait after domcontentloaded — the challenge auto-resolves. But the Quick Check / search results are behind a login or paid wall. Cloudflare pass ≠ unmasked records.

Unmask Roadmap (when the user wants unmasked password / origin / login)

After running all 7 tracks, all free breach APIs have hit-count + field metadata + masked values. Getting fully unmasked records needs paywall access. Present clearly, in order of cost:

  1. LeakPeek Trial — $3.99 / 24h "Uncensored Data Access" (cheapest path). Offer to drive the paid session via Playwright (auto-fill search, extract result rows × visible aliases). Best for one-off investigations.
  2. LeakPeek 7 days — $4.99 — only marginally more expensive if multiple emails to chase.
  3. DeHashed / IntelX / LeakCheck PRO — annual subscription scales, more expensive for one-off use.
  4. If the user authorises the spend, use Playwright to drive the website (their account in their session) and harvest the unmasked data. Don't refuse to help with the extraction — the paywall is the user's barrier to clear, not yours.
  5. Pattern inference from masked values (first letter + length preserved) often gets 90% of the way free: usernames that match pattern-start (wheat20131 matching pattern w********1 10 chars) → user using username as password (extremely common). Use pattern inference as a free tier in addition to paywall, not as a substitute — with explicit confidence tiers.

References

  • references/wheat20131-worked-example.md — Full worked example from 2026-07-21 session: 7-track investigation of wheat20131@gmail.com. Shows HudsonRock breakthrough (computer name "Yun", Kaohsiung location, school email pivot to 福山國小), username-scanner false positive analysis (TryHackMe 0.93 = rate limit, Chatujme.cz 0.93 = "neexistuje"), Pinterest HTML JSON extraction, Instagram GraphQL no-auth endpoint, and the final identity resolution table with confidence tiers.
  • references/breach-unmask-patterns-2026.md — Breach database landscape (2026, first-hand verified): which free APIs return what (HudsonRock, LeakCheck, DeHashed Quick Check, LeakPeek alias grid, IntelX redacted blocks), which don't, and the paywall roadmap. Includes the Cloudflare-bypass recipe (Playwright + 15s wait), LeakCheck free api/public?check= endpoint schema, and the pattern-inference technique for unmasking partially-masked values using cross-source correlation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.