• Rewritten from scratch to make use of what I've learned making other JBs and various prompting experiments.
• Frames the task as a "simulation" instead of "collaborative writing".
• More interactive, to always give the user something to play off of.
• More amusing writing.
• Switched from paragraph limit to word limit, to allow for more freedom in paragraph length.
• Adjusted default temperature and Top P for more creative output. (v18.1)
• Context size lowered to 24k. (v18.1)
• Split off variety instructions into separate section, and expanded. (v18.1)
• Added randomness to response length to break up repetition. (v18.1)
• Renamed sections to match new ST wording. (v18.1)
• Introduced {Markers} for prompt subsections, also now used by story/example start tokens. (v18.1)
• Moved post-history prompt to correct place, and main prompt higher up to increase its priority. (v18.1)
• Improved newline consistency. (v18.1)
• Trimmed the fluff from the JSON export. (v18.1)
• Confirmed compatibility with Sonnet "3.6". (v18.1)
• Fixed bug with ST reinserting removed Description / Persona sections. (v18.2)

• Most sections reworded to be shorter and clearer.
• Replaced most writing style guidelines for less stiff prose.
• Removed <protag>, not needed anymore with the other reinforcements.
• Number of paragraphs now tweakable via a setvar.

• Rewrote content guide again with new anti-horny techniques.
• Fixed characters knowing facts about {{user}} that were never said.
• Added XML back-references to prefill.
• Distingushed Human from {{user}} to better fight positivity bias.
• Removed a prefill line; let me know if you get more refusals now.
• Fixed bold text instruction to do it on single words rather than entire phrases.
• Moved anti-meta to prefill to strengthen it for more complex cards. (v16.1)
• Fixed anti-meta blocking post-response banter but not pre-response. (v16.2)
• Removed rule to bold stressed words; not JB's job. (v16.2)

• Used the new "Assistant Impersonation Prefill" field to fix Impersonate for good. (needs ST 1.12)
• Cleared the new "User first message" field. It was never supposed to be used.
• Replaced rependence on the [Story start] marker with a new <history> tag.
• Added a new Wrting Style section with some sane defaults.
• Clarified which sections are optional.
• Moved Jailbreak field to the very bottom of the section order.
• Minor rewordings and punctuation improvements.
• Revert adding <history> tag - not compatible with Claude 3 API. (v15.1)
• Small anti-horny improvement. (v15.2)

• Rewrote prefill and continue prompt to fix Impersonate and Continue (broken in v12).
• Rewrote content guide for better unpozzing; now more likely to generate crude content / swearing and be mean to {{user}}.
• Moved NSFW instructions from prefill to content guide.
• Reverted removal of reinforcement for not talking as {{user}}. This breaks Impersonate again. The implementation of Impersonate in ST is fundamentally broken, and needs a complete rewrite to not clash with the design of the prompt. (v14.1)

• Slimmed down Definitions section.
• Removed all remaining writing style guidelines.
• Split off explanation of the model's task into a separate section.
• Expanded on model's role to fix length creep.
• Added variety to response length. (v13.1)
• Added content guide section to reduce bias. (v13.2)

• Completely rebuilt to fix speaking for {{user}}.
• Merged Sonnet and Opus releases.
• Purged unused sections.

v11.1 (Sonnet), (Opus)

• Last major update. Consider it in maintenance mode. If you like it overall but noticed and fixed an issue, let me know and I'll integrate it.
• Removed CoT. While they do what you ask, they also always inevitably make the prose duller and less imaginative. My verdict is that CoTs are good when customized for a card, but it's not the job of a JB to have one.
• Moved anti-horny to prefill.
• Streamlined and trimmed down most sections.
• Removed tag reference from prefill (made Claude think it should generate tags in its initial responses.) (v11.1)

v10 (Sonnet), (Opus)

• Added brand new CoT, now mandatory.
  • Resolves responses getting longer over time.
  • Resolves low swipe variety on Opus.
  • Improves POV detection.
• Added reinforcement for not writing for {{user}}.
• Improved definitions.
• Simplified main prompt.
• Disabled all default preferences for Opus.
• Reduced temperature back to 0.9 for Opus.

v9 (Sonnet), (Opus)

• Opus version initial release:
  • Removed almost all writing guidance.
  • Removed writing style enforcement.
  • Added request to continue current writing style.
  • Added nudge to prevent repetition.
  • Tweaked temperature from 0.9 to 1.0.
• Removed and shortened section comments.
• Changed default context size to 40k.

• Added Arc's Hornkiller 9001 to NSFW preferences (used with permission).
• Changed listing intros to system commands.
• Improved prefill for better focus.
• Added note to maintain tense/POV to default prompt.
• Added entry to prevent metacommentary / content warnings. (v8.1)
• Strengthened anti-horny slightly with prefill mention. (v8.1)

• Compacted almost all sections.
• Removed more reiteration from prefill.
• Removed the more outlandish default style rules.
• Split basic definitions into a list.

• Added more guidelines to default writing style.
• Expanded default NSFW preferences in an attempt to keep Claude's horniness in check.
• Added rule to prevent generation of content warning OOC.
• Updated wording to prevent some unwanted outputs. (v6.1)
• Improved section comments for clarity. (v6.2)
• Enabled (empty) jailbreak section to improve card compatiblity. (v6.2)
• Reworded default style further for better effectiveness. (v6.2)

• Disabled CoT by default again.
• Added setup instructions to CoT comment.
• Removed HTML tags that hide the CoT.

• Reworked CoT to be easier to parse.
• Added event listing to CoT for more variety.
• Enabled CoT by default.
• Removed <prompt> tag, it confused the model with no extra benefit.
• Wrapped CoT formatting with triple backticks so that it's taken more literally. (v4.1)
• Adjusted default model settings for users who import and forget (perfectly valid!) (v4.2)

• Added optional CoT section (off by default).
• Fixed missing persona section.
• Improved XML tag structure.
• Fixed XML tag references to be self-closing as per Anthropic docs.
• Assigned the model a role as per Anthropic docs.

• Added comments to every section for easier tweaking.
• Strengthened prefill to get past more refusals. (Thank you kind anon!)
• Improved continue nudge to better continue from an interrupted sentence.
• Shortened standard prompt to make it more universal (now works with cards that use asterisks).
• Split up rules that shouldn't be modified from personal preferences.
• Enabled inline images, since Claude 3 supports them.