For anyone wrestling with thousands of scattered notes, research snippets, and emergent ideas, organizing that raw information into a coherent structure is a significant task. This is where concept-synthesis comes into play. It's an AI agent skill designed to help you transform vast quantities of ingested concept stubs into a clear, tiered knowledge map. The tool focuses on deduplicating and organizing these notes, providing a curated structure rather than just a dump of information.
How concept-synthesis Builds Your Intellectual Map
The skill operates through a series of five distinct phases, moving from raw data to a structured knowledge map. This process ensures that your intellectual assets are not just stored but actively managed and enriched.
The first phase is Deduplication. The agent merges duplicate concepts using a combination of Jaccard overlap for textual similarity, substring matching for exact phrases, and advanced LLM semantic analysis to identify and consolidate ideas that express the same underlying concept. This initial step cleans up your data, ensuring each distinct idea is represented once.
Next is Scoring and Tiering. Here, concept-synthesis evaluates each concept based on its frequency across your notes, the timespan over which it appears, its breadth of connection to other concepts, and its engagement (how often it's referenced or expanded upon). Concepts are then ranked into a tiered system: T1 Canon, T2 Core, T3 Support, and T4 Riff. This provides a clear hierarchy, indicating the importance and foundational nature of each idea within your personal knowledge base.
The third phase is Synthesis. For T1 and T2 concepts—your most significant ideas—the agent writes rich, detailed narratives. This involves tracing the evolution of these ideas, incorporating direct quotes from your ingested notes to provide factual grounding. The skill is designed to avoid hallucinating sources or dates, sticking strictly to the information available.
Clustering follows, where related concepts are grouped into semantic domains. The tool then builds a master map, illustrating the connections and relationships between these clusters. This gives you a high-level overview of your intellectual domains and how different ideas interconnect.
The final phase is Curation Cull. This is a reversible process where concepts are evaluated using a 6-axis rubric. Substance is weighted 2x, while packaging (e.g., clarity of original note) is weighted 1x. Based on this rubric, concepts receive verdicts: ELITE (exceptional), KEEP (valuable), MERGE (combine with another concept), or DELETE (irrelevant or redundant). This ensures that only the most valuable and relevant information remains in your curated map.
Triggers, Cadence, and Outputs
To initiate the process, you can use simple triggers such as 'synthesize my concepts' or 'build my intellectual map'. It's important to note that concept-synthesis is not a real-time tool. It runs on a cadence, typically after new notes or concepts have been ingested, allowing for a thorough and resource-intensive analysis without impacting immediate interaction.
One practical limitation of this tool is that it does not synthesize narratives for T3 and T4 concepts. This is a deliberate design choice, as synthesizing these less central concepts would be API-inefficient and detract from the focus on your core knowledge. The output of the skill is markdown files, organized under a concepts/ directory, making it easy to integrate into existing knowledge management systems. The agent interacts with your knowledge base using standard tools like query, search, and put.
Consider a researcher who has accumulated thousands of short notes, observations, and paper summaries over years. These might include raw thoughts on machine learning architectures, philosophical reflections on AI ethics, or specific programming paradigms. By triggering synthesize my concepts, the skill would deduplicate similar entries on, say, 'transformer attention mechanisms' or 'causal inference models.' It would then score these and synthesize detailed narratives for the high-tier (T1/T2) concepts, outlining their development and key points, all while linking related ideas into a comprehensive map under concepts/.
Frequently Asked Questions
Q: How often does the skill run? A: The agent runs on a cadence after concept ingestion, not in real-time. The exact timing can be configured based on your needs and the volume of new data.
Q: Does it create narratives for all my concepts? A: No, the skill specifically synthesizes rich narratives only for T1 and T2 concepts. T3 and T4 concepts are still organized and tiered but do not receive detailed narrative synthesis.
Q: What output format should I expect?
A: The skill outputs organized markdown files, which are stored within a dedicated concepts/ directory for easy access and integration.
This agent skill offers a structured approach to managing complex information. It provides a methodical way to move from disparate notes to an organized, actionable knowledge base.
