search-online - SKILL.md Agent Skill

name: search-online description: | Search Online: Research synthesis from authoritative web sources. Takes a topic query and produces a vault note synthesizing findings from recent academic papers, official documentation, and reputable blogs. Follows vault conventions with complete references, wikilinks, and folder placement—original synthesis, never copy-paste. - MANDATORY TRIGGERS: research topic, search literature, investigate topic, find papers, survey topic online, latest research

Purpose

The search-online skill researches a topic across the web and synthesizes findings into a vault note. It differs from clip-to-vault by aggregating multiple sources into a cohesive overview, rather than processing a single URL.

Workflow

Input

The agent accepts a topic query, such as:

"latest advances in retrieval-augmented generation"
"OSCP exam prerequisites and study strategies"
"LLM evaluation benchmarks 2025"
"multi-agent systems in production"
"secure code review practices"

Processing Steps

Define Search Strategy
- Primary sources (priority order):
  - Academic papers (arXiv, OpenReview, conference proceedings)
  - Official documentation and whitepapers
  - Reputable tech blogs (Anthropic, OpenAI, DeepMind, research labs)
  - Conference talks and technical talks (ACL, NeurIPS, USENIX, etc.)
- Avoid: SEO content farms, unverified claims, blogspam, outdated tutorials (>3 years)
Conduct Web Search
- Use multiple search queries to capture breadth and depth
- Include year qualifiers for recency (e.g., "2025", "2024–2025")
- Search for: seminal works, recent advances, benchmark datasets, practical implementations
Evaluate Sources
- Credibility: Check author credentials, organization, peer review status
- Recency: Prioritize sources from the last 2 years; flag older works that remain foundational
- Technical depth: Prefer sources with rigorous methodology over marketing hype
- Diversity: Balance academic papers with practical tutorials and tool documentation
Extract Key Information
- Main concepts and definitions
- Recent research findings (with numbers, benchmarks, results)
- Practical methodologies and best practices
- Open problems and limitations acknowledged by the field
- Tool ecosystem and available implementations
Search Vault for Related Content
- Query existing notes on the topic
- Identify hub notes and spoke notes to link to
- Avoid duplicating existing vault content; instead link to and expand on it
Synthesize Findings
- NOT copy-paste: Synthesize information from multiple sources into original prose
- Integrate perspectives: Where do sources agree? Where do they differ?
- Highlight key insights and novel findings from the search
- Note dates of sources and recency of research
Determine Note Type and Scope
- #note: Comprehensive overview (5+ sources, 1500+ words, multiple subtopics)
- #quicknote: Focused summary (2–4 sources, 800–1200 words, single topic)
- Add section: Search date and sources consulted (transparency about freshness)
Determine Folder Placement
- Match to vault structure (same as clip-to-vault)
- If topic is broad, consider whether a new hub note or landscape analysis is more appropriate
Add Internal Linking
- Link to existing vault notes on related or foundational topics
- Insert [[wikilinks]] strategically to connect to the vault's knowledge graph
Compile Full References
- List all consulted sources (papers, blogs, docs) with full URLs
- Indicate source type (e.g., "arXiv paper", "official documentation", "technical blog")
- Include access date and relevance to the topic
Format and Validate
- Frontmatter: Created: YYYY-MM-DD HH:MM (line 1), #note or #quicknote (line 2)
- H2 for primary sections, H3 for subsections
- Academic tone, third-person, no contractions
- Add section: Search metadata (search date, queries used, number of sources)

Output

A comprehensive vault note containing:

Opening paragraph (3–4 sentences): Topic scope, why it matters, current state of the field
Search metadata (small section): Date searched, sources consulted (count by type: papers, blogs, docs)
Main content sections: Synthesized findings from multiple sources
Key trends and insights: What's emerging, what consensus exists, what's debated
Open problems: Gaps in current knowledge or practice
Tools and resources: Implementations, benchmarks, datasets (if applicable)
Internal links: [[wikilinks]] to related vault concepts
References section: Full list of all consulted sources with types
Tags section: lowercase underscore-separated topic tags (3–6 tags)

Example Output Structure

Created: 2026-02-20 16:00
#note

Retrieval-augmented generation (RAG) has emerged as a critical pattern
for extending language models with current knowledge and domain-specific
information. This note synthesizes recent research and best practices
from 2024–2026, covering evaluation methodologies, benchmark datasets,
and production deployment considerations.

## Search Metadata

Sources consulted (February 20, 2026):
- Academic papers: 8
- Technical blogs: 5
- Official documentation: 3

## Core Concepts and Recent Advances

[Synthesized from multiple sources—not copy-paste]

## Evaluation Methodologies

...

## Open Challenges and Research Directions

...

## Tools and Implementations

...

## Related Vault Concepts

[[Retrieval Systems]], [[Vector Databases]], [[LLM Prompt Engineering]]

## References

1. [Paper title](https://arxiv.org/abs/...)
2. [Blog post](https://blog.anthropic.com/...)
3. [Documentation](https://docs.example.com/)

#### Tags

rag, retrieval_augmented_generation, llm_systems, information_retrieval

Quality Standards

Synthesis: Combine insights from 3+ sources; never regurgitate a single source
Attribution: Cite specific claims to their source (use footnotes or inline citations if needed)
Freshness: Note the search date and prioritize recent sources; flag older works as "foundational"
Balance: Acknowledge both established practices and cutting-edge research
Honesty: Highlight open problems and limitations; do not oversell consensus that does not exist
Deduplication: If this topic already exists well-covered in the vault, link to it instead of recreating

Inbox Capture Option

Research results can optionally be saved to Inbox/ as quick captures when the user prefers a fast save over full synthesis. When requested (e.g., "save search results to inbox"), create a minimal entry with:

Topic query, search date, list of key sources with URLs
1–2 sentence summary of main findings
This allows rapid capture of research; items can later be promoted to full vault notes via clip-to-vault or paper-ingest

Important Notes

Search thoroughly: Use multiple query variations to ensure comprehensive coverage
Verify recency: Include publication/update dates for all sources
Avoid SEO traps: Skip listicles, outdated rankings, and unverified claims
Link intentionally: Every wikilink should make sense in context
Add disclaimer if information is rapidly evolving (e.g., LLM benchmarks, security landscapes)