菜单
Use when converting a tabletop RPG PDF into clean, structured markdown for `corebook/`, `rpg_projects/`, or similar source-document work. Covers layout-aware extraction, multi-pass cleanup, Tales of the Old West-specific repair rules, and final normalization.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Audit and rewrite content to remove AI writing patterns ("AI-isms"). Use this skill when asked to "remove AI-isms," "clean up AI writing," "edit writing for AI patterns," "audit writing for AI tells," or "make this sound less like AI." Supports a detect-only mode, an edit-in-place mode for files, an optional voice profile (casual / professional / technical / warm / blunt), and an iterate-to-convergence pass.
Use when designing, drafting, or revising a tabletop RPG adventure, scenario, campaign chapter, or sandbox opener for any system. System-agnostic. Covers the full architecture: cold open, historical/situational frame, faction wants, NPC casting, rumor seeds, consequence tracks, scene-by-scene branching grammar, dilemma design, distributed clues, combat-encounter blocks, multi-ending reckonings, sandbox exits, and pacing rhythms. Built from the design DNA of robust, replayable, choice-driven adventures that survive contact with players. Triggered by requests to "write an adventure", "design a scenario", "outline a campaign chapter", "draft a sandbox opener", "build an introductory module", or to revise/audit an existing adventure for branching, pacing, or robustness.
Use when drafting, revising, or reviewing manuscript-facing prose for the Tales of the Old West 2E repo, or any gritty realistic western fiction in the McCarthy / McMurtry / Portis / Leonard / Hall / Lansdale / Swarthout / Hansen tradition. Applies the voice, structure, diction, knowledge base, and anti-AI rules so prose reads like native fiction — harsh, clear, physical, atmospheric, and free of generated-text tells. Covers corebook chapters, proposal prose, item and ability descriptions, flavor exposition, examples, and mixed fiction-and-rule bridges, plus standalone short fiction in the same register.
Use when evaluating game balance in tabletop roleplaying games, especially when the task requires deep analytical and psychological analysis of game theory, player expectations, GM expectations, choice pressure, fairness, lethality, spotlight balance, and perceived versus actual balance.
Use when auditing talent combinations, ability stacking, character builds, or any multi-rule interaction for balance-breaking potential. Triggers on questions like "is this combo broken", "what are the strongest builds", "does this talent stack with that ability", "can players exploit this", or any request to stress-test a rule interaction. Also use when designing new talents or abilities that might create unintended synergy with existing content. This skill distinguishes between creative high-risk play (healthy) and one-decision dominant lines (unhealthy).
Use when explaining, designing, auditing, or extending the mechanics of the Tales of the Old West 2E repo. This skill maps the rules, design space, design logic, integration points, and gameplay logic so new mechanics fit the existing engine and manuscript structure.
| name | pdf-to-rpg-markdown |
| description | Use when converting a tabletop RPG PDF into clean, structured markdown for `corebook/`, `rpg_projects/`, or similar source-document work. Covers layout-aware extraction, multi-pass cleanup, Tales of the Old West-specific repair rules, and final normalization. |
This skill documents a repeatable pipeline for turning RPG PDFs into clean markdown that can be versioned, split by chapter, and edited without losing structure.
The key idea is simple: extract conservatively, clean deterministically, and only then do any manual repair. RPG books are not plain prose. They carry running headers, decorative chapter titles, sidebars, tables, epigraphs, and column breaks that need explicit handling.
Read before you convert or repair:
SKILL.mdcorebook/AGENTS.md in the chapter or project folderFor repeatable cleanup work, prefer existing project scripts in scripts/
and project references in references/ or nearby project notes.
pymupdf4llm when layout fidelity matters and you need reading order,
table preservation, and heading detection.pdftotext -layout when the document is mostly text and you want a
fast first pass.markitdown when the file is already fairly clean and the structure is
not especially tricky.pymupdf4llm for RPG sourcebooks with headings, sidebars, and tableswordninja for dejoining spaced-out heading textInstall:
pip install pymupdf4llm wordninja
Basic extraction:
from pathlib import Path
import pymupdf4llm
md = pymupdf4llm.to_markdown(str(Path("rulebook.pdf")))
Path("rulebook.raw.md").write_text(md, encoding="utf-8")
pymupdf4llm is preferred because it is column-aware and usually keeps:
The pipeline is a sequence of independent passes. Order matters because later passes depend on earlier normalization.
Remove non-content elements that extraction often preserves faithfully:
–7–, – 88 –, or # – 14 –For Tales of the Old West, the front matter usually includes a repeated copyright/distribution block and a heavily spaced table of contents. Remove that whole region before trying to clean headings.
RPG PDFs often repeat the chapter header on every page.
Cleanup rules:
##
heading.For Tales of the Old West, this typically means Chapter N - Title should
appear once as the real section heading, not on every page.
Decorative headings often come out with spaces between every letter.
Detection rule:
Reconstruction rule:
wordninja.split() on non-trivial collapsed text.Tales-specific correction examples:
HEADING_CORRECTIONS = {
"Npcs And Abilities": "NPCs and Abilities",
"Introduction To Rpgs": "Introduction to RPGs",
"Don'T Roll Too Often": "Don't Roll Too Often",
"Npc S And Abilities": "NPCs and Abilities",
"Non Player Characters": "Non-Player Characters",
"Non Typical Harm": "Non-Typical Harm",
"Looking Back": "Looking Back",
"Looking Forward": "Looking Forward",
"F Scott Fitzgerald": "F. Scott Fitzgerald",
"Herbalist": "Herbalist",
"Bronc Buster": "Bronc Buster",
"Pistoleer": "Pistoleer",
"Rabble Rouser": "Rabble Rouser",
"Fast Actions": "Fast Actions",
"Conflict Modifications": "Conflict Modifications",
"Overwatch": "Overwatch",
"Heatstroke": "Heatstroke",
"Life in Town": "Life in Town",
"Salaries in the Old West": "Salaries in the Old West",
}
Some extractors emit image placeholders and picture-text blocks.
Rules:
<br>pymupdf4llm can detect headings but does not know the intended hierarchy.
Normalize it explicitly:
## Chapter N - Title stays ## and gets title-cased in the title portion## **bold text** becomes ### bold text## non-chapter text becomes ### text## "quote text" becomes > *"quote text"*Attribution rules:
### headings that follow a blockquote become > — NAMELong italic paragraphs often represent sidebar callouts.
PDF column breaks commonly split sentences across line breaks or blank lines.
Join text when:
Do not join structural content:
<br> inside pipe-table cells with spacesThe Tales manuscript exposed a recurring set of failure modes. Keep these rules even when the base pipeline changes.
Remove repeated page numbers, copyright/distribution lines, and the entire table of contents block when it appears as extraction noise. The TOC in this book is especially noisy because decorative spacing and dot leaders get preserved exactly.
Tales PDFs split headings across columns and pages.
Common examples from this book include:
He rba li st -> HerbalistBron C Buster -> Bronc BusterLIF E IN TO WN -> Life in TownSAL AR IE S IN THE O LD WE ST -> Salaries in the Old WestFather Brayton Car Mody -> Father Brayton CarmodyFrancisco Castellano S -> Francisco CastellanosExtraction often separates possessive s or drops apostrophes.
Man S Best Friend -> Man's Best FriendYour Horse S Tackle -> Your Horse's TackleCarson S Folly -> Carson's FollyKing S People in... -> King's People in...Use the correct apostrophe character for the manuscript and keep the fix consistent across the file.
Sometimes bold inline terms land inside heading syntax.
### line contains bold markers in the middle of sentence text, treat
it as body prose, not a real headingTwo-column layout often slices a paragraph in half and drops sidebar text in between.
Removing standalone headers is not enough.
This is the hardest class of failure.
Tell-tale signs:
Fix strategy:
Broken pipe tables need a full rewrite when the row structure is lost.
Common cases:
The rule is simple: every logical entry should end up in exactly one row.
If extraction inserts blank lines between bullet items, compact them so the list renders as a tight list instead of a loose one.
After the deterministic pipeline, chapter-by-chapter manual cleanup is still normal.
When you do manual repair for Tales:
Keep reusable notes with the project when a reference area exists:
Check the final markdown before you trust it:
<br> tags remain in body textAfter cleanup, split the single markdown file into per-chapter files only if the chapter boundaries are stable and obvious.
Typical approach:
## Chapter N - Title headings.