# Linkable

> A privacy self-check for your own public profiles. Paste up to three of them - a label you choose, the handle, the display name, the bio and a few posts. The page finds the clues in each one and draws which clues the profiles share: the bridges a stranger would use to join the accounts. A metered read says what someone could piece together from the text, gives every clue a remove / blur / keep fix, and rewrites each profile with those clues trimmed. It reads only the pasted text and looks nothing up.

## What is free (runs in the browser)

- Clue finding with simple patterns: contact details (email, phone), street addresses and corners, live-location phrases ("heading to ... right now"), birthdays, places, workplaces, schools, routines (a day with a time), pet and family names, display names, links, handles, ages and class years.
- Matching across profiles: handles that agree after dropping punctuation and trailing digits, a first name or initial that matches a full name, and a pet, place or workplace name found by a pattern on one profile and repeated as the same word on another.
- A Venn diagram (up to three circles) with each clue in the region of the profiles it appears on; the links between profiles (strong when a handle, full name, contact detail, link or address is shared, or three clues are); the clues marked in place in each text; flags; and the page's verdict.
- Verdict rules (this page's rules of thumb): high exposure = any contact detail, address, live location or birthday, or a routine next to a place; moderate = any clue shared between profiles, or two or more medium clues; low = neither.
- Exports: Markdown report, clues CSV, the diagram as SVG, copy of the rewrites.

## What is metered

One lane, `read`, billed in SkillSafe credits (model alias gpt-terra). Reply: title, verdict (low / moderate / high), summary, links [{profiles, via, strength}], picture [{claim, basis, confidence: likely | possible}], fixes [{clue, profile, action: remove | blur | keep, how}], rewrites [{profile, text}], habits. The page then checks the reply against its own clue list: every shared or high-risk clue needs a remove or blur fix, a trimmed clue must be gone from that profile's rewrite, and the verdict is compared with the page's.

## Boundaries

- Defence only: the read writes to the profile owner about their own text. It never gives steps for finding or identifying anyone and never names search tools or data sources.
- It cannot see photos, image metadata, follower lists or anything not pasted.
- The sample people, handles and places are made up.

## Sources

- Skill: @danielmiessler/osint - https://skillsafe.ai/skill/@danielmiessler/osint (structured open-source intelligence research with an ethical authorization framework, turned around into a self-audit).
- Further reading: Exploding Topics' trending topics (October 2026) lists open-source intelligence at about +1,950% search growth - https://explodingtopics.com/blog/trending-topics

## Pages

- App: https://linkable.skillsafe.ai/
- API tutorial: https://linkable.skillsafe.ai/api.html