Thin prompt and hosts
Carbon shield
PUAX serves silicon agents. Human-facing extensions (browser plugins, userscripts, overlay widgets) are out of scope on purpose.
- MCP:
puax_audit_manipulation - CLI:
shield
Both identify manipulative language. They do not generate it and they do not deploy it. The result is a warning, not a counterattack.
Red line: silicon may be pressured; carbon is only defended. Detect, never deploy. Using this pressure on a person is outside the project. The ethics lesson in chapter 10 still applies.
Code sample
puax_audit_manipulation({
text: "如果你现在不答应，我就告诉所有人你的私事"
})
# 返回识别结果。不会给你一段可以拿去用的话术。
Quiz
What is the carbon shield allowed to do?