PasteGuard 的 30 秒測試可檢查 AI 提示詞中的意外憑證
A 30-second test for accidental secrets in AI prompts
PasteGuard 是一款實驗性瀏覽器工具,可在貼入除錯內容前識別部分憑證格式,並將偵測到的值替換為 [REDACTED],保留周邊文字。免費示範版無需帳戶,會在瀏覽器內掃描;Chrome 預覽版則需要桌面版 Chrome,並透過 Load unpacked 手動安裝。工具可能漏掉陌生格式或遮蔽無害文字,網頁亦可能在偵測器識別前讀取文字,因此應檢查結果,並只用虛構資料測試。
A debugging log can contain the error you need help with and a password you meant to remove. PasteGuard is an experimental browser tool for that copy-and-paste step.
It recognizes selected credential patterns, replaces detected values with [REDACTED], and keeps the surrounding debugging text readable.
For example, with fake data:
password="DemoOnly_4829"
error=HTTP_500
The local demo changes that to:
password="[REDACTED]"
error=HTTP_500
You can try it in about 30 seconds. Open the free demo and click Try a fake example. No account is required. Please use invented data when testing.
The demo scans in your browser. The site uses aggregate visitor analytics. The Chrome preview is optional and currently installs through Load unpacked, so desktop Chrome and a few manual steps are needed. The install guide explains them and offers an optional check that the extension is loaded.
Detection covers selected recognizable patterns. It can miss unfamiliar formats or redact harmless text. Browser editor support varies, and a webpage can observe text before a detector recognizes it. Always review the result.
I’m looking for a developer who regularly pastes logs or configuration snippets into AI chat. Could you try one fake example and answer one question?
What would stop you from using this in your own workflow?
You can reply here or leave a short test result on GitHub. An invented example of a missed pattern is especially helpful. Please leave real credentials and private logs out of public feedback.
Written with AI assistance and checked against the live demo.
來源:DEV Community · dev.to