Skip to content

Liberty91LTD/cti-skills / skills/darkweb-collection

darkweb-collection

Verified

Dark web intelligence collection methodology — vendor-first access posture, sourced reference lists for 35+ underground forums and 30+ Telegram channels, OPSEC primer, passive-monitoring strategy, and bundled Python CLIs for onion-indexer search, Telegram channel monitoring, and local keyword matching. Use when the user wants to design or run dark-web collection, build a selector list, pick a vendor, or set up monitoring infrastructure.

FileKindSize
SKILL.mdskill md17.1 KB
references/access-methods.mdmarkdown12.0 KB
references/forums.mdmarkdown28.4 KB
references/opsec.mdmarkdown9.7 KB
references/passive-monitoring.mdmarkdown9.2 KB
references/telegram-channels.mdmarkdown21.3 KB
references/telegram-setup.mdmarkdown17.1 KB
scripts/keyword_match.pyexecutablescript7.6 KB
scripts/onion_search.pyexecutablescript11.8 KB
scripts/telegram_monitor.pyexecutablescript19.7 KB

references/opsec.md

View on GitHub
# OPSEC primer for dark-web collection

> Read this before doing any DIY collection. If you are using vendor feeds only, most of this is handled for you — but the persona-separation, mental-health, and incident-handling sections still apply.

OPSEC failures in dark-web collection have a long tail of consequences: deanonymisation of researchers, doxing of family members, retaliation against employer brand, criminal investigation jurisdiction issues, and (in pathological cases) physical-safety threats against analysts. None of this is theoretical — there is a public catalogue of analyst burns going back to the 2014 Silk Road takedown. Treat the rules below as load-bearing.

## Persona separation

- **Three identities, never one.** Real you ↔ work you ↔ research persona. They never share an email, a phone, a browser, a VM, a writing sample, or a timezone tell.
- **Never log into a real account from the research VM.** Not LinkedIn, not Gmail, not Slack. One slip burns the persona forever.
- **Never log into the research persona from the daily-driver host.** The browser fingerprint, the saved cookies, the autocomplete data — any one of these is enough to link.
- **Multiple personae** for different beats (one per language, one per forum tier, etc.) — never reuse a handle, never reuse a writing-style fingerprint, never reuse a profile photo (and don't use any image that ever existed elsewhere on the internet — reverse-image search is trivial).

## Hardware separation

- **Dedicated hardware** for research (a second laptop) is the gold standard. Air-gap it from your daily network when not actively collecting.
- **Failing that, dedicated VM** with full-disk encryption, snapshot rollback after each session, no shared clipboard, no shared filesystem with the host.
- **Tails OS** for amnesic ad-hoc browsing; Whonix for persistent sockpuppet sessions.
- **Never** mount research volumes on the daily host. Forensic recovery from "I'll just copy this file over" is one of the most common deanon vectors.

## Network

- **Always Tor or a research-only VPN, never your home/office IP.** Even for "just checking the leak site" — the moment you do it once unproxied, your IP is in the access logs forever.
- **Defence in depth: VPN → Tor.** Hides Tor entry from your ISP and (some) traffic-correlation attacks from passive observers. Choose a no-logs VPN paid in crypto on a separate identity if your threat model warrants.
- **Disable WebRTC** in any browser (Tor Browser does this by default). WebRTC leaks your real IP even through Tor.
- **No torrents, no streaming** from the research VM — many Tor exit nodes log abuse, and these protocols leak.

## Behavioural OPSEC

- **Timezone:** post / lurk on the *target's* operational hours, not yours. A "Russian-speaking" sockpuppet who is suspiciously active at US Eastern business hours is obviously a Western analyst.
- **Calendar:** Western public holidays (Thanksgiving, July 4th) cause inactivity in Western-run sockpuppets — schedule cover posts. Russian holidays (May 9, Jan 1-7) similarly affect Russian-speaking personae.
- **Cadence:** humans don't post in machine-regular intervals. Build natural breaks.
- **Sleep schedule:** a sockpuppet that "sleeps" 11pm–7am US Eastern leaks the analyst's geography. Either rotate handlers across timezones or don't post in real-time.

## Linguistic OPSEC

- **Machine-translation tells** are obvious to native speakers: stilted idioms, perfect grammar, overuse of a single dictionary synonym. Russian elite forums in particular have admins who flag suspected non-natives within hours.
- **Use a native-speaker collaborator** for anything beyond lurking in non-English forums. Better: hire a vetted CTI vendor who already has native-speaker analysts.
- **Slang and references:** persona-appropriate. A 19-year-old "scammer" persona doesn't drop 1990s movie quotes; a long-tenured "veteran" persona doesn't use 2024 Gen-Z slang.
- **Stylometry:** automated stylometry (writing-style fingerprinting) is a documented research field. Vary punctuation, spelling tics, sentence-length distribution, function-word frequency. The simplest defence: have your sockpuppet text edited by a different author than the analyst running it.

## Image and file OPSEC

- **Strip EXIF** from every screenshot or image before sharing. `exiftool -all= file.png` is the safe default.
- **Never post any image that exists elsewhere on the internet.** Reverse-image search (Google Lens, Yandex, TinEye) will surface the original within seconds, deanon the persona, and tell the forum admins exactly when the persona was created.
- **Profile photos:** generated faces (`thispersondoesnotexist.com`) used to be safe but are now detectable; better to use no profile photo, or a heavily edited generic icon.
- **Filenames** can leak (default Windows screenshot timestamps, default Apple "IMG_4521.HEIC", etc.). Rename before sharing.
- **PDF metadata** leaks author name + originating tool; "print to PDF" from a clean Tor Browser session is safer than uploading a Word-converted PDF.

## Counter-surveillance

- **Assume forum admins log everything.** Many forum admins are intelligence-services-adjacent; some are confirmed cooperators with state services after takedowns.
- **Assume Telegram channels can be compromised** by law enforcement or rival groups — operators have been doxed via channel ownership records leaked by ex-affiliates.
- **Assume vendors get popped.** Even commercial CTI vendors get breached or have insiders. Don't put your most sensitive selectors in vendor portals; keep them local.
- **Assume your ISP / employer / national authority logs** every site you touch. Tor + VPN are the answer; "I'll just look briefly without Tor" is not.

## Deconfliction

- **Notify your CISO + legal** before paying for invitations, interacting with sellers, attempting any active collection that could cross into "buying stolen data" or "interacting with criminals."
- **Notify the relevant national CERT / law enforcement liaison** if you find active intrusion data on your own org or a customer. Don't sit on it for a content scoop.
- **Coordinate with sector ISAC** if you find sector-wide impact (FS-ISAC, H-ISAC, etc.).
- **Never confront a threat actor directly** — even via a sockpuppet. It collapses the persona and risks retaliation.

## Persona burn handling

- **If a persona is flagged or burned** (admin call-out, sudden ban, leaked DM, suspected stylometric match): retire it. Do not try to "rehabilitate" by changing tactics. Wind down the persona's account, do not delete (deletion is itself a signal); let it go quiet.
- **Replace with a fresh persona** built from the ground up — new email, phone, browser profile, narrative, *no* overlap with the burned one.
- **Post-burn review:** what tipped them off? Stylometry? Timing? An image hit on reverse-image search? Document the lesson for the next persona.

## Mental health (load-bearing)

Sustained exposure to dark-web content — even purely cybercrime-focused, no CSAM — is documented to cause vicarious trauma in analysts. The job entails reading ransom-victim coercion, breach-victim PII, occasional accidental exposure to graphic content, and constant cynicism about the human condition.

- **Rotate analysts off dark-web duty** every 6-12 months. Long-term continuous exposure is the highest-risk pattern.
- **Provide clinical support** access. Several large CTI firms include EAP / wellness budget specifically for SOC + intel analysts.
- **Mandatory time off** after exposure to incidentally-encountered illegal-content material. Do not let a "just push through" culture take hold.
- **Buddy system:** debrief sessions weekly with a peer or supervisor. Catch creeping desensitisation early.
- **Resources:** Bellingcat have published OSINT analyst wellbeing guidance; Tech Coalition and Project VIC have CSAM-exposure-specific clinical resources for incident handlers (US-centric but applicable framework).

## Out of scope (refer elsewhere)

This skill, this OPSEC primer, and the bundled scripts do not cover:
- **CSAM (child sexual abuse material).** Discovery is mandatory-reportable in most jurisdictions. Stop reading, do not screenshot, log only the URL/identifier, and report to NCMEC (US), IWF (UK), Project Arachnid (Canada), or national equivalent. Do not investigate further; that work belongs to law enforcement and trained NGO units.
- **Active terrorism content / violent-extremist material.** Refer to national counter-terrorism units (NCTV, FBI JTTF, etc.) and to specialist platforms (GIFCT, Tech Against Terrorism).
- **Live-stream violent content.** Same as terrorism — refer to law enforcement immediately.
- **Personal-physical-safety threats** against named individuals. Refer to law enforcement and (if your org) physical security.

## A note on this skill's scripts

- `scripts/onion_search.py` queries clearnet indexers; running it from your normal IP exposes your *query string* to those indexers, which is itself a signal. Run via VPN/Tor for sensitive queries (target names, specific actor aliases).
- `scripts/telegram_monitor.py` uses your Telegram session — that account is identifiable to Telegram via the registered phone number. Treat it as a sockpuppet.
- `scripts/keyword_match.py` is local-only; no external network. Safe to run from any host.

## See also

- `access-methods.md` — vendor matrix and DIY playbook.
- `telegram-setup.md` — step-by-step setup for `telegram_monitor.py` (api_id, burner phone, session file, OPSEC ack flag).
- `passive-monitoring.md` — selector hygiene; bad selectors are an OPSEC risk too (broad selector → broad collection → broad exposure).
- `/score-source` — source rating; assume your collection sources may be compromised and rate accordingly.
- `/apply-tlp` — handle classification before sharing any collected material.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall