Getting Started

🌐 English · 中文 · 🏠 Home

Getting Started

What this skill is

Humanities Writing Companion is a Claude skill for humanities scholars whose primary deliverable is a long-form argumentative text — a journal article, a dissertation chapter, a monograph section, an essay. It is built for fields where prose IS the argument: history, philosophy, literature, cultural studies, art history, religious studies, classics, intellectual history, science studies, and adjacent humanities-aligned work.

It is not a polishing tool, not a citation manager, and not a research pipeline. It is a thinking partner that stays with you across the whole arc of a paper — from research question to submission disclosure. Generic AI writing tools pull your prose toward polishing and averaging; this skill does the opposite. It protects your scholarly voice, stress-tests your argument, and reads your draft the way a demanding humanities reviewer actually would. Its guiding principle: intellectual dialogue over surface polishing — "my hand writes my voice."

---

Install

The skill is a single plain-text file (SKILL.md) with no runtime dependencies, so it runs anywhere Claude does. Here is how to enable it in each of the three environments.

1. Claude Code (CLI)

Claude Code auto-scans ~/.claude/skills/ (user-level) and ./.claude/skills/ (project-level) on startup. Clone the repo into either location:

User-level (available in every project):

git clone https://github.com/tizzy916/claude-skill-humanities-writing-companion.git \
  ~/.claude/skills/humanities-writing-companion

chmod +x ~/.claude/skills/humanities-writing-companion/scripts/*.sh

Project-level (this repo / vault only):

git clone https://github.com/tizzy916/claude-skill-humanities-writing-companion.git \
  ./.claude/skills/humanities-writing-companion

That's it — no config step. Restart Claude Code (or start a new session), then trigger it with any phrase from First conversation below. The chmod line is only needed if you plan to use the citation helper scripts.

2. Claude Agent SDK

Load SKILL.md directly into your agent's system prompt. Because the skill is plain text with zero runtime dependencies, no special loading machinery is required — concatenate it (or the sections you need) into the system prompt and the agent gains all 11 modes.

# Pseudocode — adapt to your SDK setup
skill = open("SKILL.md").read()
system_prompt = base_instructions + "\n\n" + skill
# pass system_prompt into your Agent SDK client

Tip: SKILL.md ships with a Selective Loading Guide near the top. For a focused agent you can load only the sections a given task needs (e.g. just Four-Layer Critique + Mode B) instead of the whole file.

3. Claude.ai (web / app)

Use Claude.ai's Skills feature (Settings → Capabilities → Skills). Upload the skill folder — or at minimum SKILL.md — as a custom skill and enable it for your conversations. Once enabled, it activates the same way as in Claude Code: just start talking about your paper.

If your plan doesn't expose the Skills UI, you can still use the skill by pasting the relevant SKILL.md sections into a Project's custom instructions, or into the first message of a conversation. The behavior is identical because the skill is prompt-only by design.

---

Your first conversation

You don't run a command. You just talk to Claude about your writing, and the skill activates on natural language.

What to say to trigger it

Any of these (and many casual variants) will wake the skill:

	Trigger phrases
English	"I'm writing a paper on…", "essay", "chapter", "dissertation", "argument", "thesis", "revise", "voice", "review my section", "I'm stuck on writing", "devil's advocate", "how would a reviewer attack this?"
中文	论文 · 写作 · 改论文 · 帮我看看这一章 · 我手写我口 · 这个论证有没有问题 · 我写不下去了 · 审稿人会怎么攻击

Even a casual "take a look at this paragraph" / 帮我看看这段话 is enough.

What onboarding will ask you to set up

On first use the skill runs a short onboarding dialogue to build your writing environment. It asks six things — answer what you can; you can refine any of them later:

1. What are you writing? — title, length, and current stage (topic selection / first draft / revision / submission).
2. Discipline (routing-critical, not metadata) — collected in three layers:
   • L1 main discipline (required, pick one): Literature / History / Philosophy / Linguistics / Art studies / Religious studies.
   • L2 subfield (optional): e.g. modern Chinese literature, ethics, art history — inherits L1 and adds subfield constraints.
   • L3 cross-disciplinary (optional, can be several): e.g. intellectual history, history of science, cultural studies, digital humanities — each loads multi-discipline inheritance plus an overlay.
   • Don't see your field? The skill runs a fallback: tell it your object of study and primary method, and it infers the closest fit.
3. Citation format — Chicago/Turabian (common in history & humanities), MLA (literature & languages), APA 7th (education, some social science), GB/T 7714 (Chinese national standard), or a journal-specific template. Unsure? The skill recommends one based on your discipline and target venue.
4. Target venue — journal / conference / dissertation committee (affects format, word limits, reviewer expectations).
5. Writing language — Chinese / English / mixed, and how foreign-language sources are handled.
6. Target reader — who the paper is really written for, and their disciplinary background and theoretical position. Voice and audience are paired: the same argument needs different scaffolding for different readers.

After onboarding, the skill initializes a project folder and writes your answers to small config files (e.g. _writing-config/discipline.md, citation-style.md, style-profile.md, reader-profile.md). These persist across sessions, so next time you can just say "let's keep working on 《X》" and it picks up where you left off — no re-explaining.

A short sample first exchange

You: I'm starting a paper on how Foucault's panopticism extends
     to digital platforms. Can you help?

Companion:
  Glad to. Before we dig in, let me set up your writing environment —
  this only happens once, and you can change any of it later.

  1. What's the working title, rough length, and current stage
     (just an idea / first draft / revising)?
  2. Discipline — your main field is one of: Literature / History /
     Philosophy / Linguistics / Art studies / Religious studies?
     (Foucault-on-platforms could sit in Philosophy, or in a
     cross-disciplinary L3 like Media studies or Cultural studies —
     which feels closest to how you actually work?)
  3. Citation style — Chicago, MLA, APA 7th, GB/T 7714, or a
     specific journal's format? If you're unsure I'll suggest one.
  4. Target venue — which journal, conference, or committee?
  5. Writing language — Chinese, English, or mixed?
  6. Who's the target reader, and what's their theoretical leaning?

You: Stage: early draft. Field: Philosophy, with a Media studies
    overlay. Chicago. Targeting a continental-philosophy journal.
    English. Reader: continental theorists who know Foucault well.

Companion:
  Got it. I've recorded Philosophy (L1) + Media studies (L3 overlay),
  Chicago notes-bibliography, English, audience = Foucault-literate
  continental theorists. That last point matters: with this reader
  you don't need to *explain* the panopticon — you need to earn the
  right to extend it. Paste the paragraph you want to start with and
  I'll read it top-down: foundation first, then structure, then
  paragraph, then sentence.

Notice what the skill did not do: it didn't start rewriting. It set up the environment, flagged that your reader changes how the argument should be built, and committed to reading top-down — foundation before commas. That is what "thinking partner, not polishing tool" means in practice.

---

A taste of the four-layer read

Once you're set up, the skill's signature move is the four-layer critique — and a hard rule that it works strictly top-down, refusing to fuss with sentences while the argument underneath is broken:

🔴 Layer 1 · Foundation   "Does this paper hold up scholarly?"
🟡 Layer 2 · Structure    "How is the argument unfolding?"
🟢 Layer 3 · Paragraph    "What is this paragraph doing?"
⚪ Layer 4 · Sentence      "Is this sentence right? Well-said?"

If Layer 1 is broken, it tells you why and defers the lower layers until you've fixed the foundation. You keep cognitive agency; the skill reads for you, it does not write for you.

---

Where to go next

• Working Modes — the full tour of all 11 modes (A–K): paragraph dialogue, chapter review, drafting, devil's advocate, bottleneck help, draft revision, blind reading, research-question sharpening, literature mapping, plan-only outlining, and AI-use disclosure.
• Writing a Paper End-to-End — how the modes chain together across the lifecycle of a real paper, from research question to submission disclosure.

Welcome aboard. Tell Claude what you're writing, and let it read for you.