An LLM can't
investigate.
A trained one can.
You're right not to trust ChatGPT with casework. The difference isn't the model — it's the training. 659 skill files train Claude, GPT, Llama, or DeepSeek the way a licensed professional learned the job: chain of custody, jurisdiction, source vetting, citation discipline.
Vague summary + invented sources
Searches Google. Lists a couple of profiles that may or may not be hers. Cites a fabricated 'PeopleFinder report' that doesn't exist. No methodology, no provenance.
Methodical, cited, defensible
Runs seed-discovery-from-email → email-permutation-and-verification → alias-and-handle-pivoting → username-enumeration-across-platforms. Every finding gets a source URL, a timestamp, and a confidence grade. Stops at jurisdictional limits.
One hallucinated citation can cost you
the case, the client, or the license.
A licensed PI, journalist, fraud examiner, or attorney can't ship work that includes invented citations, missed jurisdictional rules, or no chain of custody. An LLM you merely ask to investigate ships that work routinely. An LLM trained on these skills works the methodology instead — that's the entire difference, and it's the product.
It hallucinates citations
Invents case law, statute numbers, and document IDs that sound right and aren't real.
It has no tradecraft
Skips OPSEC, chain of custody, source vetting, deconfliction. The discipline that makes findings hold up.
It doesn't know the platforms
Generic answer for LinkedIn, the same generic for Telegram. Misses the platform-specific tricks that actually work.
It treats jurisdictions the same
DMV records in California aren't the same as Texas. Wiretap law in NY isn't the same as Florida. It pretends they are.
Don't take our word.
Read a real skill.
This is the actual SKILL.md your agent loads — not marketing copy. Every skill ships as a folder: the methodology, plus runnable scripts, legal references, and field checklists.
--- name: alias-and-handle-pivoting description: Use when an authorized investigation needs to move between known and unknown handles owned by the same person, using shared bios, avatars, social graph, writing style, and posting patterns. For licensed PI, journalist source mapping, fraud, brand-abuse, attorney discovery, or authorized OSINT. --- # Alias and Handle Pivoting ## Overview People bleed identity across accounts even when they think they are compartmentalized. The discipline is to collect weak signals across many surfaces and let convergence — not any single signal — confirm attribution. ## When to Use - You have one confirmed account and suspect the subject runs others under different names. - A throwaway account is suspected to belong to a known user. - Mapping a coordinated-inauthentic-behavior cluster. - Linking a public persona to an anonymous one for journalistic accountability of a public figure. ## Core Workflow 1. Catalog every artifact attached to the known handle: bio text (verbatim), avatar (file hash + perceptual hash), banner, linked URLs, pronouns, location field, join date, follower/following lists, pinned post, language, timezone of activity, devices/clients used. 2. Avatar pivot. Compute perceptual hash (`imagehash` aHash/pHash/dHash) and exact MD5. Search reverse-image across Google Lens, Yandex, Bing, TinEye. Search the exact hash on platforms that expose CDN URLs. Many people reuse the same avatar across LinkedIn, Twitter, Reddit, Discord. 3. Bio-text pivot. Lift a 5-10 word distinctive phrase and quote-search Google, DuckDuckGo, Bing. Lift email or website fragments. Lift uncommon emoji combinations. 4. Social-graph pivot. Mutuals overlap is the strongest single signal — sort the known account's interactions by recency and look for the same set of accounts around a suspect handle. On X use advanced search `from:A @B`. On Reddit use the user-comparison heuristic: do two accounts post in the same niche subs within minutes of each other? 5. Writing-style (stylometry). Compare function-word frequency, punctuation idiosyncrasies (Oxford comma, em-dash, double space after period), typo signatures, slang, ALL-CAPS habits, sentence-length distribution. Tools: JStylo, Stylo R package; for casual use, JGAAP. Treat results as suggestive, never dispositive — adversarial obfuscation is easy. 6. Temporal-cadence pivot. Plot posting timestamps in UTC; the diurnal gap reveals timezone. Two accounts with identical sleep windows on the same days of week are correlated. Burst patterns around specific events (sports, market opens, news) further narrow. 7. Cross-platform username variants. Apply leetspeak, casing, and suffix variants from `seed-discovery-from-username` and run through Sherlock, Maigret, WhatsMyName. 8. Metadata leaks. Image EXIF (rarely intact on social), unique image dimensions or compression signatures, URL shortener accounts (bit.ly public stats), Google Doc share IDs, Calendly handle, Cash App/Venmo cashtag, Strava activities. 9. Linked-service pivots. The same email shows up on Gravatar, GitHub, Keybase; pivot via `seed-discovery-from-email`. The same Steam, Spotify, Last.fm profile if linked. 10. Build a weighted attribution matrix: rows are candidate alt accounts, columns are signal types (avatar match, bio overlap, mutuals overlap, stylometry, cadence, metadata). Require multiple independent signals before asserting same-person; record confidence as low/medium/high with evidence. 11. Preserve snapshots (archive.today, screenshots with URL and timestamp) before accounts are deleted. Route admissible attributions through `chain-of-custody-documentation`. ## Quick Reference | Signal | Tool / Method | |---|---| | Avatar | imagehash, Yandex, TinEye | | Bio phrase | Google verbatim, DuckDuckGo | | Mutuals | X advanced search, Reddit user pages | | Stylometry | JStylo, Stylo (R), JGAAP | | Cadence | timestamp plotting, sleep-window analysis | | Handle variants | Sherlock, Maigret, WhatsMyName | | Linked accounts | Gravatar, Keybase, GitHub, Steam | | Archive | archive.today, Wayback, screenshots | ## Common Mistakes - Single-signal attribution. Same avatar alone is not proof — image theft and copies are common. - Confirmation bias. Once you suspect an alt, every coincidence looks confirming. Pre-register what would falsify the hypothesis. - Ignoring base rates. In a niche subreddit, posting in the same threads is not informative; in a generic sub, it is. - Failing to preserve. Subjects often delete within hours of noticing scraping. - Mis-attributing shared-team accounts (PR firms, assistants, partners) to a single person. ## Ethical / Legal Limits Doxxing or unmasking a private individual for harassment is prohibited and may be criminal under cyberstalking statutes. Unmasking is defensible for public figures acting in public capacity, accountability journalism, fraud investigation under retained counsel, or court-ordered discovery. Platform ToS may forbid creating sockpuppet research accounts; some jurisdictions treat sockpuppet creation against ToS as a CFAA gray area (Van Buren narrowed but did not eliminate this). Document the engagement basis and route findings through `chain-of-custody-documentation`.
9 bundles. Seven craft volumes,
one Complete Library, one free.
Buy a volume once and own it — two volumes take 10% off, three or more take 15%. Add the $10/mo Update Subscription to keep everything you own current as the tradecraft, law, and platforms change.
OSINT Foundations
Find the thread. Pull it methodically.
Digital Forensics Pro
Acquire it correctly. Document every step.
Legal & Procedure
Built around the rules of evidence.
Surveillance & Tradecraft
The discipline behind the camera.
Financial Crimes & Crypto
Follow every wallet, every wire.
Specialty Investigations
From arson to cold case to cybercrime.
Analyst & Cloud Suite
Make the picture make sense.
Investigator Starter
Four headline workflows. Four domains. Free, forever.
Built for people whose work has to hold up.
If a fabricated source could lose your case, your client, your beat, or your license — these skills exist for you.
Licensed PIs
Skip tracing, asset tracing, surveillance ops, evidence-grade reporting — all built around your state's disclosure rules and DPPA/FCRA compliance.
Recommended volumeInvestigative journalists
Source verification, geolocation, document forensics, leak vetting, pre-publication review. Bellingcat-style methodology baked in.
Recommended volumeFraud examiners + AML
BSA/SAR/CTR workflows, forensic accounting, crypto tracing across BTC/ETH/Solana/mixers, sanctions and PEP screening.
Recommended volumeAttorneys + paralegals
Discovery review, FOIA, subpoena drafting, evidence admissibility, deposition prep, witness statement format, Brady/Giglio compliance.
Recommended volumeDefensive / corporate intel
Threat assessment (workplace, school, executive), insider threat indicators, brand protection, kidnap-ransom corporate response.
Recommended volumeCode enforcement + reg
Administrative inspection authority, abatement orders, OSHA workplace safety, environmental enforcement, food-safety protocol.
Recommended volumeDefensive SOC / DFIR
Disk imaging with chain of custody, memory analysis, browser artifact analysis, cloud-platform (AWS / Azure / GCP / M365) evidence collection.
Recommended volumeCounter-disinfo + OSINT
Bot/CIB detection, narrative laundering tracing, deepfake political content attribution, coordinated amplification analysis.
Recommended volumeThree steps. Then your agent acts like a pro.
Skills are plain Markdown — readable by you, loadable by Claude Code, OpenAI Custom GPTs, Llama, DeepSeek, and any agent that supports system prompts or skill files.
Buy your stack
Stripe checkout — volumes, add-ons, and updates in one payment. You get a license key, ZIP downloads, and local Claude Code plugins.
Install into your agent
One command for Claude Code. Drag-and-drop for Custom GPTs. Drop into ~/.skills/ for Llama / DeepSeek. Docs cover all four.
Ask. Watch the method appear.
Your agent now invokes the right skill at the right moment. Cited findings, jurisdictional caveats, confidence language, the whole craft.
Two upgrades when you're ready to go further.
The skills stand alone. These two add-ons multiply them — both one-time purchases.
Orchestrator.md — run a whole team of agents at once.
One file that spins up parallel specialists — people, businesses, case facts, report — and merges them into one defensible case file. 4 hours of prompting becomes 30 minutes. Free with the Complete Library.
Investigator's MCP Toolkit — the agent runs the tools itself.
36 pre-vetted MCP servers wired to the skill library: live WHOIS, username sweeps across 2,500+ sites, Wayback diffs, on-chain tracing, memory forensics. The agent stops suggesting and starts doing.
Don't ship work an LLM guessed at.
Start with 4 skills free. Build your stack when your agent earns it.
