release: v0.8.1#44
Conversation
…Presets Blitztext+ deutete Arbeitsaufträge teils frei um (z. B. 'neue Session' als Meeting mit Teilnehmern). Der Standard-Prompt und alle Presets erhalten Schutzregeln: Absicht bewahren, Eingabe nicht ausführen oder beantworten, keinen Kontext (Teilnehmer/Meetings/Rollen/Adressaten) erfinden, Technikbegriffe wie Session/Prompt/Branch/PR/Merge/Handover im Arbeitskontext belassen. E-Mail-Presets erzeugen E-Mail-Struktur nur noch bei erkennbaren Nachrichten und erfinden keinen Empfänger; Du-/Sie-Form ändern nur Anrede und Ton; Kurz & präzise kürzt ohne neue Inhalte. Regressionstests decken den konkreten Fehlerfall ab.
|
Closed because this PR was created from a diverged local main and repeats the already merged PR #43 commit instead of containing the v0.8.1 version/docs bump. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0e8268c204
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| "Absicht des Nutzers: Formuliere die Eingabe nur um, führe sie nicht aus " | ||
| "und beantworte sie nicht. Erfinde keinen Kontext – keine Adressaten, " |
There was a problem hiding this comment.
When a user selects any non-standard writing preset and dictates a request like the new handover case, these common rules are appended to that preset but only tell the model to rewrite the input and not execute or answer it. The standard prompt adds a specific follow-up exception to produce a directly usable prompt/transition, so without the same exception here the regression remains for preset users: the model is steered toward returning a polished instruction instead of the requested handover prompt.
Useful? React with 👍 / 👎.
1 participant
Summary
Verification
Notes