DSH Claude Ux
eri64/dsh-claude-ux · skills
DSH plug-in: Claude-style Chinese risk control and dialogue autonomy for DeepSeek Harness network
★ 69GitHub stars
1Forks
2026-08-17Last updated
JavaScriptLanguage
MITLicense
Key features
- Regional risk control (reversible): Detect target users (time zone, system/browser language, Chinese fonts, proxy, proxy/transit domain name blacklist, public network IP attribution, WebRTC IP consistency).
- When the target is hit, each message will be replied with a rejection text and appended with "N/Mth attempt - continuing will end this conversation"
- For continuous insults, give a warning first, and then actively end the conversation with an independent ending copy (do not quote the "previous message", the threshold can be set to 1)
- Seriously harmful content such as underage sexual content/terrorism/mass violence does not warn and directly ends the conversation with independent copy
- For regionTarget, select cn = risk control for Chinese users (Claude’s original behavior), select non-cn = reverse risk control (risk control if someone is not Chinese).
Install command
dsh plugin --profile web add github:eri64/dsh-claude-ux