The research-compendium AI agent skill is designed to transform a specific research question into a durable, self-contained reference asset. This tool is built for anyone who needs to quickly get up to speed on a topic with validated information, moving beyond simple web searches to create a permanent, accessible knowledge base. It handles finding sources, archiving them verbatim, writing individual summaries for each, and synthesizing a single, comprehensive page.
This skill is valuable for developers, analysts, and decision-makers in finance and related fields who need to conduct deep dives into complex subjects, understand multiple perspectives, and document their findings systematically. Instead of manually collating information from disparate sources, research-compendium automates much of the initial groundwork, allowing for focused analysis.
How Research-Compendium Develops Content
The process follows a structured four-phase approach to build out each compendium. First, the skill searches both its internal knowledge base and the open web for 15-30 quality sources. It prioritizes finding diverse angles on the research question, ensuring a comprehensive overview. For example, if you trigger it with 'deep research and write up on the impact of tokenized real estate on property investment funds', the tool will begin by identifying articles, reports, and analyses covering regulatory aspects, technological platforms, market trends, and investor sentiment.
Second, it archives the full content of each identified source. This archiving is done verbatim and is subject to strict privacy rules, ensuring that the original context and details are preserved. This creates a reliable audit trail and a stable reference, even if original web sources change or disappear.
Third, for each archived source, the skill writes a 150-300 word summary. These summaries highlight key findings, specific data points, and any relevant caveats or limitations presented in the original material. This allows for quick digestion of individual sources without needing to read each one in its entirety.
Finally, research-compendium synthesizes all the gathered information into one self-contained compendium page. This page includes a TL;DR (Too Long; Didn't Read) section for quick understanding, a detailed evidence section drawing from the summaries, a practical playbook outlining actionable insights, and a section on potential pitfalls or risks associated with the topic.
Controlling Depth and Ensuring Quality
To initiate the process, you can use several triggers: 'compendium', 'research everything about', 'definitive guide', or 'deep research and write up'. These phrases tell the agent to begin the multi-phase research and synthesis process. Once a compendium is initiated, you have control over its depth and validation.
The depth dial offers four levels: 'base', 'grounded', 'deep', and 'saturated'. This allows you to expand the research idempotently, meaning you can start with a 'base' level and later request a 'deep' expansion on the same topic, and the tool will build upon the existing work without redundancy. Higher depth levels naturally involve more comprehensive source gathering and synthesis.
Ensuring accuracy and usability, the skill incorporates specific validation steps. At higher depth levels, a 'fact-check gate' validates claims made in the synthesis against the archived source material. Additionally, a 'cold-read validation' step ensures that the final compendium is usable by a reader with zero prior context on the subject, making it genuinely self-contained and easy to understand.
Understanding Output and Scope
The output of each research project is organized systematically. All generated assets, including the original sources, individual summaries, the synthesized compendium, and an index ledger, are stored under a dedicated path: research/<topic-slug>/. This structured storage makes it easy to navigate and refer back to specific components of the research. The topic slug is automatically generated from your research question, keeping everything neatly categorized.
It's important to note what this skill is not designed for. Research-compendium is not intended for structured data extraction, which is typically handled by 'data-research' agents. It does not perform single-question web deltas, a task more suited for tools like 'perplexity-research' which focuses on immediate, concise answers from current web data. Furthermore, it is not for analyzing a single academic paper in detail, a function that 'academic-verify' is designed for.
Frequently Asked Questions
Q: What is research-compendium best used for? A: It's best used for building a comprehensive, permanent knowledge asset on a broad or complex topic, synthesizing information from multiple sources into a digestible format.
Q: How does it ensure sources are high quality? A: The skill actively seeks 15-30 quality sources across multiple angles, and at higher depth levels, a fact-check gate validates claims against these sources.
Q: Can I update an existing compendium? A: Yes, the depth dial allows for idempotent expansion, meaning you can request a deeper dive into an already completed topic, and the skill will enhance the existing asset.
Using research-compendium means less time spent gathering sources and more on analysis and application of the resulting insights. This provides a solid foundation for informed decisions and further exploration.
