[iamwavecut]

Summary

• Last-message note. It asks for a text reply that starts with the substance, without a reaction, rating, agreement, acknowledgment, interjection, name or lead-in. It also lists the opening words of the bot's last eight replies so the model avoids them. A custom persona gets no list, the same rule as for re-sampling.
• shared_core.
  • No name or greeting up front.
  • No short lead-in sentence before the point.
  • Past replies are context, not a template.
  • The anti-repeat window matches the code guard (eight replies).
  • Explicit translation requests always use translate_text.
  • Strong emotions on a message go to react_to_message.
• system_contract. Native tool calls come first, and the final text has no report of the tool call. turn_boundary and tool_contract state the reaction exception.
• Tool descriptions. draw_image and generate_song no longer suggest text beside the call; the session never published that text. react_to_message says that text returned with it is delivered.
• Re-sample note. For a repeated opening, the note now asks for a different first-sentence construction without quoting the rejected words.
• Session. Text returned together with react_to_message alone is the final reply. The session delivers it without a second model call. The text passes the same leak guard and reply-shape review as any final answer. Text beside any other tool call is still dropped.

Prompt replay

The replay used 131 production requests without tool calls and 20 requests with tool calls, all with untrimmed history. The new prompt was compared with the deployed one.

	deployed	this PR
short intro sentence before the answer	39%	22%
reaction openers	11%	8%
opening repeats one of the last 8 replies	3.1%	1.6%
copies of history markup	2/131	1/131
native tool calls	8/20	12/20
executable tool calls	19/20	19/20

On the 16 replies whose opening repeated a recent one, the re-sample started the same way in 7 cases with the old note. With the new note it did so in 1 case.

Checks

• cargo fmt --all
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test -p openplotva-dialog -p openplotva-llm -p openplotva-prompts -p openplotva-app

[github-actions]

	Warnings
⚠️	Prompt files are cache-sensitive contracts; confirm prompt shape and role markers were preserved intentionally.

	Messages
📖	Danger checked 9 changed file(s).

Generated by 🚫 dangerJS against f25dcef

[iamwavecut]

Danger prompt warning: the prompt changes are intentional and keep the existing shape. system.prompt and _shared_core.prompt keep every tag and the partial; only the text inside the tags changed. last_message_wrapper.prompt keeps the element, the attributes and the {{{message}}} slot. The note text changes and gains an optional {{recent_openings}} list. resample_note.prompt changes two repeated_opening sentences. None of these files has role markers.