Overriding a scan warning

Saving a page returns a warning instead of a saved bookmark when the injection scanner flags its content, or when the page has no extractable text (an unrendered JavaScript shell).

{
  "status": "scan_warning",
  "risk_score": 0.85,
  "signals": ["direct_override"],
  "matches": ["...ignore all previous instructions and..."]
}

Deciding

• content_unscannable — the daemon couldn’t read the page, not that it’s malicious. Common on JS-heavy sites. Safe to override; the extension’s browser-captured HTML usually gives a better result than force alone.
• A real signal with a low score (~0.7) — often a false positive on an article that quotes an injection example or discusses prompt security. Read the matches snippet and decide.
• A high score (0.85+), system_token, or exfiltration — the page is trying to manipulate an AI reader. Don’t save it. If you must, know that the chunk metadata will carry the risk score into every search result.

Overriding

Re-save with force=true:

POST /collections/{id}/bookmarks?force=true

This skips the fetch-and-scan gate entirely and creates the bookmark immediately. The background index job still runs and still records per-chunk risk metadata.