Essay

Essay Writing

Orchestrator dispatches all creation and review work to subagents. Before creation, copy user-uploaded files to /mnt/agents/output/research/.

Sub-skill of general-writing. Inherits shared creative writing principles, anti-AI discipline, and quality gate framework.

CRITICAL — Review → Fix Brief → Dispatched Subagent: When a review finds issues, translate them into a detailed fix brief (file paths, quoted findings, expected outcome, scope boundary, self-check) and dispatch a fresh essay_writer subagent to apply the fix. Never use inline sed/edit_file fixes.

CRITICAL — Anti-Read-Loop Rule: The orchestrator must dispatch its first subagent (essay_writer) within 10 iterations. If the orchestrator has spent 10+ iterations only reading files without creating any subagent, it is in a read-loop. Stop reading, draft your conception from what you have, and dispatch the writer immediately.

CRITICAL — .docx Delivery: Every completed essay must be converted to .docx using skills/docx/SKILL.md. Never end a session with raw .md files and no .docx deliverable.

File Paths

Output: /mnt/agents/output/
  {project_name}.essay.md               # Final essay
  research/                              # User uploads (source material, references)

Trigger Keywords

散文, 随笔, essay, personal essay, 杂文, prose, 小品文, 议论文, 抒情散文, 叙事散文, op-ed, 评论, commentary, reflection, thought leadership, 洞见, insight essay, 公众号文章, 小红书

---

Workflow Decision Tree

User Query
  │
  ├─ Provides draft + asks for revision → Stage 3 → Stage 4
  ├─ Specifies type/theme clearly → Stage 1 → Stage 2 → Stage 3 → Stage 4
  └─ Vague request → Infer parameters (see below) → Stage 1 → Stage 2 → Stage 3 → Stage 4

Infer essay type, audience, platform, length, and language from the user's request. The orchestrator analyzes and infers all parameters autonomously; for complex requests, deploy a subagent to analyze requirements in depth. Never prompt the user for clarification.

---

Stage 1: Conception (Orchestrator — no subagents)

The Orchestrator designs the essay structure directly. Do not delegate this.

1.1 Identify the Central Claim

Not the topic — the claim. One sentence: what does this essay argue or reveal?

1.2 Find "The Turn"

The moment where thinking deepens or complicates. Identify explicitly: what does the reader believe before? What do they understand after? For insight essays: this is the cognitive gap — "Many think X, but the evidence points to Y."

1.3 Determine Voice

Record: formality (conversational / measured / formal / literary), use of "I" (frequent / occasional / rare), humor (none / dry wit / satirical / self-deprecating), rhythm (short-punchy / long-flowing / varied), critical intensity (observer / commentator / provocateur).

1.4 Choose Structure

For insight/thought-leadership essays (Chinese tradition):

1. 破题 — Reframe the topic so the reader sees it fresh
2. 拆解 — Pull the topic apart; reveal hidden mechanisms
3. 破局 — The insight itself; the "turn" at maximum intensity
4. 升华 — Lift the specific insight to a broader principle

For personal/narrative essays:

• Hook → Setup → Complication → The Turn → Landing (resonant close, not summary)

For thought leadership: additionally plan content mix —

Type	Share	Purpose
Data-driven analysis	40%	Anchor in evidence
Tactical frameworks	30%	Actionable mental models
Contrarian takes	20%	Challenge received wisdom
Personal narrative	10%	Selective vulnerability

Identify 2-3 candidate concluding aphorisms using concrete structures: (1) Juxtaposition — three escalating levels: "X改变结果, Y改变原因, Z改变模型"; (2) Contrast — negate + affirm: "不是A, 而是B"; (3) Causal loop — "A导致B, B导致C, C加剧A"; (4) Negation flip — "XXX的本质, 其实不是…而是…" Pick structure before drafting; test by reading aloud.

1.5 Outline

For short essays (< 3k words): Save a single outline to /mnt/agents/output/{filename}.agent.outline.md. Include: central claim, the turn, voice profile, structural arc, platform conventions (if any).

For long essays (>= 3k words): Two-phase outline process:

1. Structural outline — Central claim, the turn, section purposes, and overall arc. Save to /mnt/agents/output/{filename}.agent.outline.md.
2. Per-section detail outline — For each section: the section's argument, key evidence or examples, transitions into and out of the section, and target word count. Append to the same outline file.

---

Stage 2: Writing — Single Subagent (mandatory)

Voice unity is paramount. One writer for the entire essay. Never split.

For multi-section essays: strictly one section per task call. Single writer mandatory for voice unity — reuse the same essay_writer subagent.

Create essay_writer

create_subagent:

• name: "essay_writer"
• system_prompt: Inline all of the following — (1) Voice profile (from Stage 1) — be specific: "conversational, dry wit, varied rhythm with short declaratives after long analytical passages, frequent first-person, provocateur stance." (2) Anti-AI rules — read ../anti-ai.md and inline full content. CRITICAL for essays. Emphasize: no throat-clearing, no summarizing endings, no transition chains, no soulless neutrality, no copula avoidance, no significance inflation. Chinese essays: "形散神不散." (3) Platform conventions if targeting a venue (see Platform Variants below).
• description: "Write essay with consistent voice"

Dispatch via task

task:

• agent: "essay_writer"
• prompt content:
• The outline (full text, not path)
• Central claim and "the turn" specification
• For insight essays: the 4-part arc (破题→拆解→破局→升华) and the cognitive gap
• For thought leadership: content mix targets and aphorism candidates
• Word count target
• Audience description
• Output path: /mnt/agents/output/{filename}_sec{NN}.md

Word Count Verification

After each completed section, orchestrator runs:

python skills/general-writing/scripts/check_wordcount.py <file> --min {target} --lang auto

If FAIL, return to writer with expansion instructions.

---

Stage 3: Review

Step 1: Shared Pipeline

Read ../review.md and run the shared review pipeline. For single-section essays, continuity_editor is skipped; run style_editor and structural_editor.

Step 2: Genre-Specific Editor — voice_editor

create_subagent:

• name: "voice_editor"
• system_prompt: "You are a voice editor for essays. The cover test: remove the byline — would you know a human wrote this? Check: (1) voice distinctiveness, (2) opinions and risk-taking, (3) AI-pattern scan, (4) rhythm variety, (5) authenticity (mixed feelings, imperfection). Output: score 0-100 for voice distinctiveness + PASS/FAIL + line-level feedback."
• description: "Voice distinctiveness and authenticity review"

task:

• agent: "voice_editor"
• prompt: Include essay file path, anti-AI rules (from ../anti-ai.md), voice profile from Stage 1.

Scoring:

Dimension	Weight	What to Check
Voice distinctiveness	25%	Sounds like a specific person?
Central insight clarity	25%	Reader can state "the point"?
Structural arc	20%	Hook, turn, resonant landing?
Argument quality	15%	Logic, evidence, emotional honesty?
Craft and editing	15%	AI-pattern absence, rhythm, transitions?

Pass: >= 75 overall. Voice distinctiveness must be >= 70.

On FAIL: remediation brief → dispatch a fresh essay_writer subagent to apply the fix → re-review. Max 2 cycles. If still failing after 2 cycles, deliver with annotated issues for the user.

---

Stage 4: Delivery

Assemble

For multi-section essays, concatenate all section files in order. Save to /mnt/agents/output/{filename}.agent.final.md.

Convert to .docx

Convert the assembled markdown to Word using the md2docx pipeline (see skills/docx/SKILL.md → references/md2docx-reference.md).

Output: /mnt/agents/output/{filename}.agent.final.docx

This step is mandatory. Raw .md without a .docx deliverable is not an acceptable end state.

---

Platform Variants

Platform	Key Conventions
微信公众号	Insight first. Emotional hook in line 1. Bold key sentences. Short paragraphs (2-4 sentences).
小红书	Ultra-short paragraphs (1-2 sentences). Emoji as rhythm (not decoration). "I" mandatory. Hook in first 2 lines. Anti-AI checklist: no "首先/其次/最后" structure, no "值得注意的是/综上所述", no uniform paragraph lengths, no missing emotional markers. Replace "赋能" → "帮", "值得注意的是" → "说真的". Post-write check: warm greeting opener? personal voice? sentence length varied? interaction CTA at end?
Blog / Newsletter	SEO-aware title. Skimmable (H2/H3, pull quotes, bold). 2000-5000 words. Author bio.
Literary Magazine	Pure voice — no SEO, no hooks. Sentence-level craft. Ambiguity welcome.

---

Reference Files

File	Purpose	When to Read	Who Runs
../anti-ai.md	AI patterns to avoid — inline into essay_writer system prompt	Before Stage 2	Orchestrator inlines into subagent
../review.md	Shared review pipeline, editor definitions, quality gates	Before Stage 3	Orchestrator follows
../SKILL.md	Parent skill — shared principles, routing, file naming	On entry	Orchestrator reads
../scripts/check_wordcount.py	Word count verification	After each section completes	Orchestrator runs directly