Labs · Card Workbench Pro
The card editor that
diagnoses while you write.
Form editors check structure. This bench checks the writing: impersonation hits highlighted inline, AI-slop and AI-flavor scores per field, token budget bars, and craft guidance from real card-writing playbooks — running live as you type. Plus system_prompt editing, multi-member group-chat pack assembly, and IndexedDB drafts that fit megabyte cards. Nothing leaves your machine.
Drop a card to start editing
PNG / JSON / .charx · up to 64MB · nothing uploads
Parsing, editing, diagnostics and exports all run in your browser. Files and drafts never upload.
Drafts on this device
No drafts yet. Edits auto-save into this browser's IndexedDB — large cards fit fine.
Solo card: create or drop a file and the diagnostics rail runs while you type. Group pack: bundle several cards into one distributable group-chat payload.
Where the diagnostics come from
Every engine in the rail ships elsewhere on this site as a standalone tool, and they're all ports of the rules used in the Foreverse production line: the structural checkup is the app import chain's parser; impersonation detection is the engine behind our impersonation checker, built on subject-position analysis (not just spotting the word 'you'); English slop scoring uses a lexicon calibrated on 30 popular community cards; the Chinese AI-flavor ruleset is the v10.7 browser port of what audits our official cards. The workbench's contribution is running them continuously against your edit state and mapping every hit back to the exact field and sentence.
The group-chat assembly mirrors the production pack validator rule for rule — member count band, roster-vs-card name cross-check, director length in code points, avatar size caps — so the exported payload is shaped exactly like what the publishing pipeline validates.
What it doesn't do
It measures signals, not quality — a card can pass every check and still be boring, and a great card can trip a rule on purpose. The zh flavor ruleset only scores mostly-Chinese text and the English lexicon only judges English prose; outside those ranges the rail abstains and says so. Token counts are estimates, not a tokenizer.
Group-chat export is the payload layer only: publishing a pack (manifest, content hash, review) happens in the app's author center. .charx imports fine but exports as JSON/PNG — the web can't repack its asset bundle. And drafts live in this browser's IndexedDB: export the file when it matters.
FAQ
How is this different from the Card Studio editor?
Card Studio edits structure and runs the structural checkup (dead lorebook entries, broken macros, token thresholds). The Workbench keeps all of that — same parser, same lorebook editor — and adds what no form editor has: a live content-diagnostics rail. While you type it runs impersonation detection (does your card speak for {{user}}?), AI-slop scoring for English prose, the zh AI-flavor ruleset for Chinese prose, per-field token budget bars, and craft guidance from real card-writing playbooks. It also edits system_prompt / post_history_instructions, assembles group-chat packs, and auto-saves multiple drafts to IndexedDB, so multi-megabyte cards fit.
Is anything uploaded while I edit?
No. Parsing, every diagnostic engine, macro preview, PNG embedding and zip packing all run in your browser. Drafts save to this browser's IndexedDB on your machine — not to an account, not to a server.
What do the inline highlights mean?
Each field has a Marks tab: red marks are impersonation hard hits (the card speaks, acts or feels for the user), yellow are weak signals (scene-teleport openers, perception hijacks), purple are stock slop phrases from the English lexicon, orange are Chinese cliché hits. Hover any mark for the rule name. The rules are the same engines behind our standalone impersonation checker and slop doctor — not a second opinion.
What exactly does the group-chat assembly export?
A payload zip in the exact shape the Foreverse community group-chat pack (group_chat) expects: payload/group_chat.json with version/members/strategy/director, one card JSON per member under payload/members/, optional compressed avatars under assets/avatars/. Honest boundary: this is the payload layer, not a publishable .fvpack — the manifest, content hash and review happen in the app's author center. The workbench mirrors the production validator's limits (2–12 members, roster names must match in-card names, director under 2,400 code points), so what you export passes that gate without rework.
Are the token counts exact?
They're character-ratio estimates, the same estimator used across our card tools — good for spotting which field is overweight, not a tokenizer. Real models differ by 20–30%. The English slop lexicon only judges English prose and the zh flavor ruleset only scores mostly-Chinese text; when a card is out of a ruleset's calibrated range, the rail says so instead of faking a clean score.