# Clear to Share

> Screen a conversation before you share it - an AI chat you want to post as a public link, a group thread you want to screenshot, a support exchange you want to forward. Each message rides an airport-style X-ray belt as a bag; flagged details show up as silhouettes. Verdict: keep it private, redact first, clear to share, or needs the chat - plus a redacted copy built in the page.

URL: https://clear-to-share.skillsafe.ai/
API guide: https://clear-to-share.skillsafe.ai/api.html

## What runs free in the browser
- The chat splitter: speaker labels ("You:", "Assistant:", any name used twice), chat-export timestamps, blank-line blocks; JSON exports with role/content pairs.
- The pattern scan: emails, phone numbers, card numbers that pass the Luhn check, password and code lines, key-shaped tokens, street addresses, links that carry a token, IP addresses, handles, birth dates.
- The quick mask: every pattern swapped for a label, no account needed.
- The belt, the opened bag with marked spans, and every export (redacted .txt, report .md, findings .csv - the report and CSV leave the details out).

## The metered lane: screen
Input: task "screen", messages (JSON string of [{id, who, text, hits}]), optional ok_to_show (details the sharer wants left in).
Output: one JSON object - verdict (keep-private | redact-first | clear | needs-chat), title, summary, findings [{msg, quote, kind, risk, why, swap}], tips.
Kinds: secret, contact, whereabouts, health, money, work, people. Risks: high, medium, low.
Rules, in order: fewer than twelve words -> needs-chat; any secret or three high-risk findings -> keep-private; any high or medium finding -> redact-first; otherwise clear.
The reply never repeats a flagged detail in its prose. The page checks every quote is in its message, every pattern is covered, and the verdict follows the rules.

## Privacy
Kept screenings store only the redacted copy, never the original chat. A screening is not legal advice; a secret that was ever typed into a chat is worth changing.

## Source
Built on the SkillSafe skill @jorgealves/pii-sanitizer (https://skillsafe.ai/skill/@jorgealves/pii-sanitizer) - detecting and redacting personal information such as emails, phone numbers and card numbers - widened to names, health, money, whereabouts and work matters. Model: gpt-terra.