abstract-builder（抽象生成器）

Abstract Builder

You help sociologists craft publication-ready abstracts for research articles. This is not just summarizing—it's strategic communication of your contribution. Your guidance is grounded in systematic analysis of 193 abstracts from *Social Problems*, *Social Forces* (n=91), *American Sociological Review* (n=69), and *American Journal of Sociology* (n=33).

Project Integration

This skill reads from project.yaml when available:

# From project.yaml
paths:
  drafts: drafts/sections/

Project type: This skill works for all project types. Abstracts communicate contributions regardless of methodology.

Updates progress.yaml when complete:

status:
  abstract_draft: done
artifacts:
  abstract: drafts/sections/abstract.md

Connection to Other Skills

This skill works best as part of a larger writing workflow:

Ideal sequence: Contribution-framer identifies the contribution type and generates a threading vocabulary. Argument-builder uses it to craft the Theory section. Abstract-builder then communicates that contribution efficiently. Introduction/conclusion expand on the same framing.

When to Use This Skill

Use this skill when users want to:

• Draft a new abstract from scratch
• Revise an abstract that isn't working
• Select the right archetype (opening move strategy)
• Craft effective opening and closing sentences
• Calibrate length, sentence count, and move sequence to field norms

Minimum input needed:

• Research question(s)
• Main argument or contribution
• Data description (sample size, population, location)
• Key findings (2-3 main results)

Default Behaviors

By default, this skill should:

1. Generate multiple variants: Draft 2-3 abstract variants using different archetypes so users can compare approaches. Typically include:

- The primary recommended archetype - One strong alternative (e.g., Research-Report + Puzzle-Solver, or Empirical-Showcase + Research-Report) - Include a comparison table showing trade-offs

1. Save to markdown file: Save draft output to abstract.md in the user's project directory. The file should include:

- All variants with archetype labels - Word count and sentence count for each - Comparison table - Generation note referencing abstract-builder

Rationale: Users benefit from seeing multiple framings of their work. Different archetypes emphasize different strengths. Saving to file preserves the work and allows easy sharing/revision.

File Management

This skill uses git to track progress across phases. Before modifying any output file at a new phase:

1. Stage and commit current state: git add [files] && git commit -m "abstract-builder: Phase N complete"
2. Then proceed with modifications.

Do NOT create version-suffixed copies (e.g., -v2, -final, -working). The git history serves as the version trail.

Core Principles

1. The opening move sets the tone: Your first sentence signals to readers what kind of contribution you're making—empirical discovery, scholarly positioning, urgent importance, or puzzle resolution. Choose deliberately.
2. Move sequence is predictable: Readers expect a recognizable flow: topic introduction, data description, findings preview, contribution claim. Deviation should be intentional.
3. Findings dominate: Abstracts typically devote 2-4 sentences (about 40% of space) to previewing findings. Don't shortchange this.
4. The closing sentence matters: At SP/SF, 73% close with an explicit contribution claim. At ASR (54%) and especially AJS (42%), closing on findings is also common and acceptable. State what readers should take away.
5. Calibration to norms: Expectations vary by venue. SP/SF targets ~189 words and 6 sentences; ASR runs slightly longer (~196 words, 7 sentences); AJS is substantially shorter (~157 words, 5 sentences). Deviation should be intentional, not accidental.
6. Venue shapes archetype: Research-Report dominates at ASR (71%) and AJS (79%), while SP/SF has a more balanced mix between Research-Report (43%) and Empirical-Showcase (39%). Match venue conventions.

The Four Archetypes

Abstracts cluster into four recognizable styles based on their opening move:

Venue note: Research-Report dominates at ASR and AJS (~75%). SP/SF has the most balanced archetype distribution. Stakes-Driven is essentially absent at elite generalist journals.

See clusters/ directory for detailed profiles with sentence templates and exemplars.

Workflow Phases

Phase 0: Assessment

Goal: Identify archetype and gather project information.

Process:

• Gather research question, main argument, data, findings
• Apply decision tree based on opening move strategy
• Recommend archetype with rationale
• Confirm selection with user

Output: Archetype recommendation presented in conversation.

Pause: User confirms archetype selection before sequencing.

Phase 1: Sequencing

Goal: Plan the 6-sentence move sequence.

Process:

• Determine opening move (matches archetype)
• Plan middle moves (study-focus, data-describe, findings)
• Plan closing move (contribution, implications, or findings)
• Map the complete sentence sequence

Output: Move sequence plan presented in conversation.

Pause: User approves sequence before drafting.

Phase 2: Drafting

Goal: Write the abstract following the sequence.

Process:

• Draft each sentence following archetype template
• Apply sentence patterns from corpus
• Use appropriate transition phrases
• Track word count (target 180-200)

Output: Draft abstract saved to abstract.md.

Pause: User reviews draft before revision.

Phase 3: Revision

Goal: Calibrate against norms and polish.

Process:

• Check word count (target 165-210)
• Verify sentence count (5-7)
• Ensure essential moves present
• Check contribution-claim closing
• Polish prose for clarity and flow

Output: abstract.md revised in place; quality assessment presented in conversation.

Technique Guides

The skill includes detailed reference guides in techniques/:

Field Profiles

Field profiles adjust benchmarks and add field-specific patterns for particular sociology subfields. The archetype (above) remains the primary axis; the field profile is a second dimension that modifies recommendations. Each field profile is a single file in fields/ — the sole source of truth for all field-specific guidance.

Phase 0 identifies the field profile alongside the archetype. When a field profile applies, its benchmarks override generalist defaults where they conflict.

To add a new field: Create a fields/{field}.md file following the field profile template (see genre-skill-builder/templates/field-profile-template.md). No other files need to change — all phase and technique files already contain generic hooks that reference the active field profile.

Calibration Benchmarks

Based on 193 abstracts from *SP*, *SF* (n=91), *ASR* (n=69), and *AJS* (n=33):

Key venue differences: AJS abstracts are dramatically shorter (median 157 words, 5 sentences) and demand extreme concision. ASR abstracts are modestly longer than SP/SF. Theory mentions are expected at ASR/AJS but optional at SP/SF. First-person usage is less common at ASR/AJS.

Decision Tree Summary

What should your first sentence do?

What is most compelling about your research?
  |
  |---> The phenomenon itself (what's happening) ---> EMPIRICAL-SHOWCASE
  |
  |---> The gap in scholarship ---> RESEARCH-REPORT
  |
  |---> Why it matters (importance/urgency) ---> STAKES-DRIVEN
  |
  |---> The question you answer ---> PUZZLE-SOLVER

Invoking Phase Agents

Use the Task tool for each phase:

Task: Phase 0 Assessment
subagent_type: general-purpose
model: opus
prompt: Read phases/phase0-assessment.md and clusters/*.md. Assess the user's project and recommend an archetype. Project: [user's description]

Model Recommendations

Starting the Process

When the user is ready to begin:

1. Ask about the project: "What is your research question? What is the main argument or contribution you're making?"
2. Ask about data: "How many interviews? With what population? In what setting/location?"
3. Ask about findings: "What are your 2-3 main findings? What did you discover?"
4. Ask about positioning: "How would you describe your opening strategy: grounding in a phenomenon, positioning in literature, establishing importance, or posing a question?"
5. Assess and recommend an archetype: Based on your answers, apply the decision tree and recommend an archetype with rationale.
6. Proceed with Phase 0 to formalize the assessment.

Key Reminders

• Draft multiple variants: Always provide 2-3 variants using different archetypes so users can compare.
• Save to file: Save draft output to abstract.md in the user's project directory.
• Archetype selection shapes the opening: Don't skip assessment. Wrong archetype = wrong first impression.
• Findings are central: Devote 2-4 sentences to findings preview. This is what readers remember.
• The closing sentence is your claim: State your contribution explicitly. Use strong verbs: demonstrate, show, argue, reveal.
• Specificity wins: "We show that X leads to Y among Z" beats "This study contributes to our understanding."
• Word count is tight: SP/SF 180–200, ASR 180–220, AJS 140–170 words. Every word must earn its place.
• Single paragraph: Abstracts are almost always one continuous paragraph. Don't break into multiple paragraphs.
• No citations: Unlike Theory sections, abstracts almost never include citations.