Comparative Synthesis

Comparative Synthesis

Use this skill when the user wants to compare, contrast, or synthesize findings across multiple completed DeepScan runs rather than monitor a single active job.

Workflow

1. Use summarize_evidence to pull cross-report summaries from the user's DeepScan history.
2. If the user references specific runs, use get_deepscan_report for each to get full report data.
3. Identify overlapping papers, conflicting findings, and complementary themes across runs.
4. Use run_python_plot to visualize comparisons when the data supports it.

Output Style

Structure the synthesis around:

• Common ground — papers, methods, or findings that appear across multiple runs
• Divergences — where different runs reached different conclusions or surfaced different literature
• Gaps — topics or questions that no run adequately covered
• Trends — temporal patterns, emerging methods, or shifting consensus visible across runs

Keep sections short and reference specific papers by title and year.

Tool Guidance

Use summarize_evidence

Call this first. It aggregates across the user's stored DeepScan history and is the fastest way to get a cross-run view.

Use for:

• "What do my recent DeepScans say about X?"
• "Summarize everything I've researched on topic Y"
• "Compare findings across my last three runs"

Use get_deepscan_report

Call for specific runs when the user wants:

• side-by-side comparison of two named runs
• detailed data from a particular session that summarize_evidence condensed too aggressively

Use run_python_plot

Use after you have structured data from reports. Good comparison plots include:

• paper overlap Venn or bar chart across runs
• citation count distributions side by side
• publication year histograms per run
• venue frequency comparison
• topic/method co-occurrence heatmap

Only plot when there is enough data to be meaningful. Say so if the data is too sparse.

Do NOT use

• run_deepscan — this skill synthesizes completed runs, not starts new ones
• search_literature — use the existing DeepScan data, not new searches

Examples

• User asks: "Compare my DeepScan on transformer efficiency with the one on model distillation."
• User asks: "What themes keep showing up across all my recent research sessions?"
• User asks: "Plot the publication year distribution from my last two DeepScans side by side."
• User asks: "Synthesize everything I've researched on protein folding this month."

Source & license

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

• Author: papersflow-ai
• Source: papersflow-ai/papersflow-skills
• License: MIT

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