docs: drop the claim that prompts are never logged

Upstream logs caught error messages unsanitized regardless of the debug
switch, and a JSON.parse failure quotes an excerpt of the request body.

Helix-Harness: claude-code
Helix-Model: claude-opus-5-5
This commit is contained in:
helix-b committed 2026-10-06 21:55:06 +00:00
1 parent 5603a42be0
commit 4f8e5e1adb
5 files changed
+10 -8

No files matched your search

+1 -1
View File
@@ -122,7 +122,7 @@ Sets the API key and the verbose-logging switch.
- **Cost:** the service restarts, since the proxy reads its config at start.
- **Repeat safety:** idempotent. **Leaving the key blank keeps the existing one** rather than clearing it, so the action can be used to toggle logging without re-entering the key.
- **The key is never echoed back into the form.** The logging toggle is pre-filled; the key is not.
- **Verbose logging** adds a log line naming the model of each chat request, and logs errors that cut off a streamed reply. Prompt and reply contents are never logged.
- **Verbose logging** adds model information for accepted chat requests and logs OpenAI stream errors. Error messages may include sensitive content, even when verbose logging is off.
- **The key's shape is checked before it is written.** Upstream's status page rejects a malformed key outright, but a bad key written straight into the config file would only surface later as a failed request — so the same check is applied here.
**Requests are billed to whichever key is set.**