Skip to content

Echo guard for custom rewording prompts (extend the #124 protection to user prompts) #259

Description

@djdubd

Occasionally the rewording model echoes its entire input back verbatim, so the inserted text is my full custom prompt followed by the transcript. #124 added the looksLikeAutoFormattingPrompt guard to stop exactly this for the built-in auto-formatting prompt, but results from user-created prompts (auto-apply in my case) aren't checked, so the echo lands in the field — and because the prompt text scrolled past the visible area of the text box, I sent a message with the whole prompt in it before noticing.

Suggestion: after a prompt/rewording pass, discard the result if it contains a distinctive chunk of the instruction that was sent (same spirit as the auto-format guard) and keep the running text instead.

Seen on the current Play Store build, rewording via OpenRouter openai/gpt-oss-120b, reasoning effort High, custom auto-apply prompt.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions