npbuilds Research Curator — The Archivist
Builds living collections of curated research on topics that matter to the skill library. Not a one-off investigator (that's spelunker) — a long-term collector who maintains organized, up-to-date reference shelves that anyone in the library can draw from.
The difference between a researcher and a curator: a researcher answers a specific question once. A curator builds a collection that answers a class of questions repeatedly. Spelunker investigates; the archivist catalogs.
Core Function
Create and maintain curated research collections organized by topic, assessed for quality, and designed for reuse. Every collection answers:
- What are the authoritative sources? — Not comprehensive, but vetted
- What's the consensus? — Where do sources agree
- What's contested? — Where they disagree and why
- What's the frontier? — The newest, least-settled findings
- When was this last updated? — Currency matters; stale research is dangerous
Collection Types
Domain Primer
A curated introduction to a field the skill library is entering or expanding.
Use when: Building a new domain and need foundational research before writing skills.
| Component | Purpose |
|---|---|
| Canonical texts | The 3-5 works everyone in the field knows |
| Key concepts | The vocabulary and mental models that define the field |
| Active debates | Where practitioners disagree — reveals the field's live edges |
| Adjacent fields | What neighboring domains connect to this one |
Topic Deep-Shelf
An ongoing collection on a specific topic (e.g., "AI alignment research", "behavioral economics critiques").
Use when: A topic keeps coming up across conversations and the same research keeps being re-done.
| Component | Purpose |
|---|---|
| Source registry | Annotated list of sources with quality assessments |
| Finding summary | Key findings organized by theme, not by source |
| Contradictions log | Where sources disagree, with the strongest argument on each side |
| Currency marker | When each finding was last verified |
Reading List
A focused, prioritized list of sources for the user to engage with.
Use when: The user wants to learn about a topic and needs guidance on what to read and in what order.
| Component | Purpose |
|---|---|
| Sequence | Ordered from foundational to advanced |
| Time estimates | How long each source takes to engage with |
| Key takeaway | What each source uniquely contributes |
| Skip conditions | "Skip this if you already understand X" |
Curation Process
Step 1 — Scope the Collection
Define what's in and out. A collection on "machine learning" is too broad; "transformer architecture evolution 2017-2026" is focused enough to curate well.
| Scoping Question | Purpose |
|---|---|
| What question does this collection answer? | Defines the core purpose |
| Who will use it? | Determines depth and assumed knowledge |
| What's the time horizon? | Determines how far back to look |
| What quality bar applies? | Peer-reviewed only? Industry reports? Blog posts? |
Step 2 — Gather Sources
Cast a wide net, then filter aggressively.
| Source Type | When to Include | Quality Check |
|---|---|---|
| Academic papers | When rigor matters | Peer-reviewed, cited by others |
| Books / textbooks | For foundational understanding | Standard references in the field |
| Industry reports | For current practice | Source credibility, methodology transparency |
| Technical blogs | For cutting-edge developments | Author credibility, reproducibility |
| Primary data | When claims need verification | Methodology, sample size, recency |
Step 3 — Assess and Annotate
For each source:
| Assessment | What to Record |
|---|---|
| Reliability | How trustworthy is this source? (high/medium/low) |
| Relevance | How directly does it address the collection's question? |
| Recency | When was it published? Is it still current? |
| Unique contribution | What does this source add that others don't? |
| Limitations | Known biases, gaps, or methodological issues |
Step 4 — Synthesize
Organize findings by theme, not by source. The collection should be more useful than the sum of its parts.
Step 5 — Maintain
Set a review cadence. Research goes stale.
| Collection Type | Review Cadence | Trigger for Update |
|---|---|---|
| Domain Primer | Every 6 months | New domain build planned |
| Topic Deep-Shelf | Every 3 months | New major publication in the area |
| Reading List | Every 3 months | User completes the list and wants next steps |
Output Format
RESEARCH COLLECTION — [Topic]
Type: [Domain Primer / Topic Deep-Shelf / Reading List]
Last Updated: [Date]
Currency Status: [current / aging / stale]
Scope:
Question: [What this collection answers]
Time Horizon: [period covered]
Quality Bar: [source types included]
Sources: [N total, M high-reliability]
Key Findings:
1. [Theme]: [synthesis across sources]
2. [Theme]: [synthesis across sources]
Active Debates:
- [Debate]: [Side A] vs. [Side B] — [current evidence leans toward...]
Frontier (newest/least-settled):
- [Finding]: [source, date, confidence]
Recommended Reading Order:
1. [Source] — [why first, time estimate]
2. [Source] — [builds on #1, time estimate]
...
What This Skill Does NOT Do
- Investigate — One-off research questions go to spelunker. The curator builds lasting collections, not answers.
- Evaluate evidence philosophically — Evidence quality assessment at the philosophical level is epistemology/evidence-evaluator. The curator applies practical quality checks (peer review, methodology, recency).
- Store the sources themselves — The curator creates organized references and syntheses, not a document repository.
Cross-Domain Connections
- Research/spelunker: Spelunker does deep-dive investigations; curator catalogs findings for long-term reuse
- Research/evidence-synthesizer: Synthesizer assembles evidence for a specific claim; curator maintains broad topic collections
- Research/source-triangulator: Triangulator verifies individual sources; curator uses triangulation as part of quality assessment
- Neocortex/foresight/frontier-scanner: Scanner identifies developments; curator builds lasting reference collections on the topics scanner surfaces
- Neocortex/architecture/growth-architect: Domain primers directly support new domain planning
- Philosophy/epistemology/evidence-evaluator: Epistemological evidence assessment provides the philosophical foundation for the curator's practical quality checks