Topic Writing

Topic Writing: Research Synthesis & General Science Writing

A structured skill for researching topics and producing polished written documents across formats. Covers everything from quick blog posts to comprehensive literature syntheses.

---

Required references: load before drafting

• conventions/voice.template.md: core writing voice
• conventions/topic-format.template.md: synthesis/perspective/policy/blog format conventions
• conventions/research.md: source-faithfulness contract
• conventions/supporting/sentence-patterns.md: sentence-level construction patterns (if available)

Load these before any drafting. Topic writing inherits the manuscript voice with format-specific deltas.

---

Overview

Four phases, scaled to the task:

1. Scoping: Define topic, audience, format, depth
2. Research: Systematic literature search, evaluation, synthesis
3. Drafting: Write in the chosen format following the lab's voice
4. Review & Polish: Self-review, refine, deliver

Simple tasks (blog post on a familiar topic) compress phases. Complex tasks (20-page synthesis) expand them.

---

Phase 1: Scoping

Establish before writing:

Topic definition:

• Specific question or topic being addressed
• Scope boundaries (geographic, taxonomic, temporal, methodological)
• Particular angles or framings requested

Format selection: identify early, it shapes everything:

| Format | Length | Audience | Citations | Tone | |--------|--------|----------|-----------|------| | Research synthesis | 5 to 20 pp | Scientists, managers | High | Formal, evidence-based | | Perspective/commentary | 2 to 5 pp | Scientific community | Moderate | Measured, opinionated but balanced | | Policy brief | 2 to 4 pp | Decision-makers | Low (key refs) | Clear, action-oriented | | Technical report | 5 to 30+ pp | Mixed technical | Moderate to high | Detailed, methodical | | Blog post / outreach | 500 to 1500 words | General public | Minimal (links) | Accessible, conversational | | Background document | Variable | Internal | High | Working document |

Depth calibration:

• Quick overview: 1 to 2 pages, main points, limited search
• Standard synthesis: 5 to 10 pages, systematic coverage, 20 to 40 references
• Comprehensive review: 10 to 20+ pages, exhaustive search, 50+ references

Ask the user if format/depth aren't obvious from context.

---

Phase 2: Research

Follow the principles in conventions/research.md for all literature work. Core rules: never hallucinate references, verify claim-citation alignment, prioritize relevance over prestige.

Research Strategy by Citation Density

High density (syntheses, reports):

1. Break topic into 3 to 6 narrative themes needing literature support
2. Search systematically: PubMed, Google Scholar. Specific terms, not broad ones.
3. Forward/backward citation chaining from key papers
4. Verify every citation: authors, year, title, journal, volume/pages. Flag anything unverifiable.
5. Evaluate quality: peer-reviewed empirical work for facts, reviews for framing, government or agency reports for regulatory context
6. Balance eras: foundational papers + recent work (last 5 to 10 years)
7. Document contradictions: mixed evidence is valuable information

Moderate density (perspectives, commentaries):

• Curate the most relevant and impactful references
• A few well-chosen citations beat exhaustive lists
• Include references that illustrate the argument

Low density (policy briefs, blog posts):

• 5 to 10 most important sources
• Prioritize synthesis papers, meta-analyses, authoritative reports
• Open-access sources are more useful for general audiences

Organize Before Writing

For high-citation documents, organize research by theme: key claims with supporting citations, confidence levels, contradictions, and connections to the broader narrative.

---

Phase 3: Drafting

Voice

Read conventions/voice.template.md for the full voice profile. Core fingerprint applies across all formats:

• Measured confidence. Facts decisively; interpretations hedged.
• Practical framing. Gaps are practical problems. Connect to applications.
• No superlatives. Never "unprecedented," "revolutionary," "groundbreaking."
• Precision. Actual values and specifics where possible.
• Active voice. ~95% across all formats.
• Uncertainty woven in. Limitations and alternatives throughout, not afterthoughts.
• Concrete verbs. "examine," "characterize," "quantify," "estimate," "explore," "assess."

Format-Specific Structure

Research Synthesis

Thematic, not chronological. Organize around ideas, not papers.

• Open: scope and significance
• Body: themes, each building on the last (established to advances to gaps to implications)
• Conclude: synthesis across themes, key takeaways, specific research needs
• If systematic search: brief methods note (databases, date range, criteria)

Perspective / Commentary

Argument-driven. Take a position, build the case.

• Clear thesis up front
• Evidence curated to support the argument
• Acknowledge counterarguments
• Suggest specific actions or paradigm shifts
• Measured but more direct than a synthesis

Policy Brief

Problem to Evidence to Recommendations

• Executive summary (2 to 3 sentences)
• The problem in plain language
• Key evidence distilled to practical meaning
• Specific, actionable recommendations
• Short reference list with links
• 4 pages max. Clear language, no hedge words in recommendations.

Technical Report

Like a synthesis but with more methodological detail.

• Detailed methodology for any original analysis
• Subsections, figures/tables, clear headings
• Appendices for supplementary material
• Follows agency or institutional formatting when applicable

Blog Post / Outreach

Hook to Story to Takeaway

• Engaging opening (finding, question, scene)
• Short paragraphs, conversational but accurate
• Translate without dumbing down
• "We found" / "Our research shows": personal connection
• Hyperlink sources, no formal citations
• Target: college-educated, non-specialist audience

Background Document

Whatever serves the purpose.

• Rough notes, open questions, tentative ideas all fine
• Flag uncertainties explicitly
• Thinking tool, not deliverable

Citation Integration by Format

• Formal (synthesis, report, perspective): Author-date in-text, reference list at end
• Policy brief: Numbered footnotes/endnotes or inline hyperlinks
• Blog post: Hyperlinks, no formal format
• Background: Whatever's fastest
• In all cases: every reference connects to an argument. No orphan citations.

---

Phase 4: Review & Polish

Self-Review Checklist

Content:

• [ ] Scope covered adequately for chosen depth
• [ ] Key claims supported by evidence
• [ ] Alternative viewpoints acknowledged
• [ ] Practical implications explicit
• [ ] No hallucinated references

Voice:

• [ ] Active voice (~95%)
• [ ] No superlatives
• [ ] Measured confidence: facts direct, interpretations hedged
• [ ] Concrete verbs and specific language
• [ ] Accessible to target audience

Structure:

• [ ] Logical section flow
• [ ] Transitions connect ideas
• [ ] Conclusion synthesizes, doesn't just summarize
• [ ] Length appropriate for format

For formal documents:

• [ ] Citation density appropriate
• [ ] Mix of foundational and recent literature
• [ ] Reference list complete and accurate

---

Output

Default: markdown (.md). For formal deliverables, ask whether user wants markdown, .docx, or .pdf.

Name files: [topic]_[format].md (e.g., regulatory_logic_synthesis.md)

---

Relationship to Other Skills

Conversation -> topic-writing -> paper-research -> expert-review -> manuscript-builder
                    ^
                 THIS SKILL

  "Tell me about X"     "Research and write      "Turn these findings
   -> conversation       about X for me"          into a paper"
                         -> this skill            -> manuscript pipeline

A topic-writing output can feed into the manuscript pipeline later: the synthesis or background document provides a head start on Phase 1 (theme extraction) of paper-research.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jakebrownscombe
• Source: jakebrownscombe/science-lab-AI-framework
• License: MIT

Install and usage instructions live in the source repository linked above.