- Home
- Skills
- Bdambrosio
- Cognitive Workbench
- Synthesize
synthesize_skill
- Python
9
GitHub Stars
2
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill bdambrosio/cognitive_workbench --skill synthesize- Skill.md4.0 KB
- tool.py23.9 KB
Overview
This skill integrates content across multiple documents or between two documents to produce new understanding. It combines, compares, and distills information from Collections to surface themes, contrasts, and integrative conclusions. Use it to create narratives, structured comparisons, or executive and technical summaries from aggregated source material.
How this skill works
The skill flattens Collection items and applies optional focus filtering to keep only relevant chunks. It uses a hierarchical map-reduce approach for long inputs (auto-chunking around ~16k characters) and can compare two inputs when an explicit second Note or Collection is provided. Output format options control structure and compression: narrative, comparison (JSON), executive, technical, or comprehensive, and a custom instruction can override defaults.
When to use it
- Identify themes and trends across a set of research papers or notes
- Compare methodologies, results, or claims between two documents or Collections
- Produce an executive summary or technical synthesis from many sources
- Generate a structured JSON comparison highlighting shared and unique points
- Aggregate item-level extractions into a coherent report
Best practices
- Provide a clear focus string to guide relevance filtering when you need targeted synthesis
- Choose format based on audience: executive for leaders, technical for practitioners, comprehensive to preserve nuance
- Set compression_ratio to control brevity vs. detail for narrative/technical/comprehensive outputs
- When comparing, always supply the 'other' parameter and use format="comparison" to get structured results
- Pre-run a map(extract) step if you need consistent per-item facts before synthesis
Example use cases
- Synthesize dominant research trends from a conference’s accepted papers into a 400-word executive overview
- Compare two competing system designs and produce a JSON report showing shared themes and contradictions
- Create a technical synthesis of methodological evolution across several studies with moderate compression
- Integrate product feedback notes into a narrative prioritizing recurring usability issues
- Aggregate per-article extractions into a comprehensive review preserving nuance for each source
FAQ
The operation fails: target is required and must be non-empty.
Can I control output length?
Yes—set compression_ratio for narrative/technical/comprehensive formats; executive outputs aim for 300–500 words by default.