Curated, not exhaustive. Trusted, not abundant.
The Research/ folder is the agent's ground-truth context. It is not a list of links. It is a curated set of source material the project actually relies on.
If you can't write a one-line note explaining why a source is in the folder, the source doesn't belong in the folder.
That rule alone solves 80% of Research/ problems.
Five categories:
- Primary sources — original research papers, books, named-publication articles, primary documents
- Verified quotes with citations — a quote-pack file with attribution, date, source URL, and rhetorical purpose
- Domain best practices — well-regarded methodology guides for the deliverable type
- The user's own seed sources — references the user already trusts
- Recent news headlines (when relevant) — pulled from named publications, dated, with URLs
- General "background reading" — the agent has training data
- Quote-aggregator sites (azquotes, brainyquote, quotefancy) — they repeat each other and many quotes are paraphrases or misattributions
- Vendor marketing content unless explicitly requested
- Anything the user wouldn't actually trust
- Files the project doesn't actually use
When in doubt: leave it out. A small high-signal folder beats a large noisy one.
The folder has a Research/README.md that indexes everything. Format:
| File | Source type | Why it's here |
|---|---|---|
| machlup-1962.pdf | Book | Coined "half-life of knowledge" — used on Slide 2 |
| ieee-spectrum-1966-technical-obsolescence.pdf | Magazine article | Original 35-year/10-year engineering half-life numbers |
| nyt-1991-engineer-supply-affects-america.pdf | Newspaper | Updated half-life to <5 yrs engineering, <3 yrs software |
| kruchten-2008-half-life.pdf | Magazine article | 2008 update on software-engineering half-life |Each row is a single source with a one-line rationale. If the rationale is hard to write, the source doesn't belong.
For projects that include attributed quotes (slides especially), maintain quote-pack.md separate from the rest of the research. Format:
## Q1 · Pasteur — preparation creates the surface area for luck
> "In the field of observation, **chance favours only the prepared mind.**"
**Source.** Louis Pasteur, inaugural address as Professor and Dean of the new Faculté des Sciences, Université de Lille, 7 December 1854.
**Why it lands here.** *(rhetorical purpose for the project)*
**URL or full citation for verification.** *(link or page reference)*Rule: no quote ships in a deliverable that isn't in quote-pack.md. This single discipline eliminates almost all attribution errors.
A primary source is the originator of a claim, fact, or quote. A secondary source cites the primary.
Example:
- Primary: the original 1925 R. A. Fisher paper introducing inferential statistics
- Secondary: a 2023 textbook discussing Fisher's framework
- Tertiary: a Medium post citing the textbook
For the Research/ folder, prefer primary. When primary is unavailable, secondary is acceptable if you note it explicitly. Tertiary should not appear unless flagged as a research lead, not a citation.
This sounds pedantic. It isn't. The single most common failure mode in agent-generated work is citing a quote-aggregator site as if it were authoritative. The discipline catches that.
- A new factual question came up that the existing sources don't answer
- A claim in the work needs a stronger source than what's already in the folder
- The user mentions a source they trust that wasn't added at kickoff
When not to add sources:
- The agent found something interesting but it's not actually used in the work
- Background reading the user might want someday but doesn't need now
- Tertiary sources
If you find yourself adding sources you won't actually use, you're hoarding, not curating.
For projects that reference current events (a talk, a memo, a market analysis), keep current-news headlines separate from durable research:
Research/
├── README.md
├── primary-sources/
│ ├── (durable research)
│ └── ...
├── headlines/
│ ├── headlines-pack.md # ranked list of recent headlines with publication, date, URL
│ └── ...
└── quote-pack.md
Headlines have a half-life of weeks. Primary sources have a half-life of years. Don't mix them.
After multiple projects in similar domains, you'll notice the same primary sources recurring. Promote them.
Create a ~/research-library/ directory in your home folder with subdirectories per domain (statistics/, economics/, presentation-design/, etc.). Symlink (or copy) recurring sources into the project's Research/ folder rather than re-downloading.
Over time, the library becomes valuable in its own right — your personal canon of trusted sources, curated and indexed.
A well-curated Research/ folder for a typical knowledge-work project has:
- 5-12 items total
- 1-3 of the user's own seed sources
- 3-7 primary sources (papers, books, primary documents)
- 1-3 best-practice references
- A
quote-pack.mdif quotes are involved - A
headlines-pack.mdif current events are involved - A
README.mdindex where every entry has a one-line rationale
If you're under 5 items, you may have under-researched. If you're over 15, you may be hoarding. Curate, don't accumulate.