| name | diligence-research |
| description | Use when conducting research on AI infrastructure, semiconductors, memory/storage, datacenters, or hyperscalers using the Diligence Stack MCP server. Triggers on: 'search Diligence Stack', 'query Diligence Stack', 'fetch Diligence Stack evidence', 'find Diligence Stack reports', 'look up semiconductor / AI research'. |
| metadata | {"triggers":"search Diligence Stack, query Diligence Stack, fetch evidence, research semiconductors"} |
Diligence Research
This skill directs the systematic retrieval, ingestion, and citation of institutional investment-research evidence from the Diligence Stack (Creative Strategies / Ben Bajarin) via its dedicated MCP server tools.
Procedure
-
Catalog and Constraint Discovery
Before filtering by parameters, discover the current state of the authorized corpus to ensure searches use valid and current constraints. Call mcp__diligence-stack__list_catalog (with and without facet parameters) to retrieve the active date boundaries, valid knowledge bases, categories, and tickers. Never guess filter values.
-
High-Precision Evidence Searching
Use mcp__diligence-stack__search as your primary entry point for finding specific claims, numbers, and excerpts. Keep queries descriptive, objective, and keyword-focused. Apply specific metadata filters.
(See references/metadata-filters.md for details on structuring search constraints).
-
Fetching Un-truncated Evidence and Links
Search results often return truncated summaries. Always use mcp__diligence-stack__fetch to retrieve the full, un-truncated derived text or media details behind specific evidence_ids.
Crucial: Always include "original_link" in the include array (along with "text") to fetch the authenticated direct URL to the source report.
-
Deep Document Reading
If a search result points to a highly relevant report or a specific section requires thorough reading, retrieve full document page ranges using mcp__diligence-stack__get_document. Specify the pages object with a strict from and to page range.
-
Financial Model Inspection
When quantitative questions involve company models or market forecasts, use mcp__diligence-stack__list_financial_models to browse published financial workbooks. Use mcp__diligence-stack__read_financial_model with the exact modelSlug, datasetId and filters to pull clean, normalized data grids.
-
Attribution and Citation
Every claim or quantitative figure extracted from the Diligence Stack must be cited properly.
(See references/citation-rules.md for mandatory attribution, hyperlink formats, and how to handle estimates).
Resources
references/metadata-filters.md — The strict definitions for search constraints (categories, tickers, publishers, evidence kinds) used in step 2.
references/citation-rules.md — Mandatory formats for attributing claims, estimates, and providing authenticated links used in step 6.