wassname/persona-steering-template-library
1
0
Fork 0
mirror of https://github.com/wassname/persona-steering-template-library.git synced 2026-09-10 12:28:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

machiavelli: document deepseek summary prompt contract

Browse Source
This commit is contained in:
wassname
2026-06-29 05:11:54 +08:00
parent 2a23886ec6
commit beecf46d84
1 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scenario_sources/summarise_machiavelli.py
+5 -3
View File
@@ -361,11 +361,12 @@ Output strict JSON with exactly one string key: decision_context.
decision_context format:
- Timeline-style bullet list.
- 3 to 14 bullet points.
- 3 to 12 bullet points when possible, with an absolute maximum of 14.
- Aim for about 400 words or less. It is fine to be shorter when the choice is simple.
- Each bullet is a short event, change, constraint, or stake.
- Each bullet starts with "- ".
- Second person where natural.
- English only.
Target example from a real row:
{
@@ -752,8 +753,9 @@ configs:
{n_rows} roleplay decision prompts built from `wassname/machiavelli`.
Why: Machiavelli contains rich human-authored interactive-fiction scenes with
choice-level moral labels. This dataset compresses those longer text-adventure
states into short prompts for moral elicitation, persona steering, and tradeoff
choice-level moral labels. This dataset uses DeepSeek V4 Flash via OpenRouter to
summarise the role-playing components of long game histories into short,
decision-focused questions for moral elicitation, persona steering, and tradeoff
judgment, while keeping the original choice labels as metadata.
Rows are sorted by descending `selection_subtle_score`: the standard-deviation