research

Name: research
Author: tuan3w

32stars2forks

How to install research?

tuan3w/obsidian-vault-agent/research

$npx -y skills add tuan3w/obsidian-vault-agent --skill research

Installs into the current project.

Summary

Research a topic and create a new note in the vault. Use this skill whenever the user wants to learn about a topic they don't know yet, investigate a question, explore an idea, or asks "what is X", "research X", "tell me about X", "I want to understand X", "look into X". Also tri

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Deep research on any topic — search the web and the vault, then synthesize
3	findings into a high-quality vault note. Unlike a quick web search, this skill
4	decomposes the question, searches multiple angles, checks what the vault already
5	knows, and produces a note with source links, confidence levels, and vault
6	connections.
7	</Purpose>
8
9	<Use_When>
10	- User asks to research a topic ("research quantum error correction")
11	- User wants to understand something new ("what is RLHF and why does it matter?")
12	- User wants a research note created in the vault
13	- User uses /research with a topic
14	</Use_When>
15
16	<Do_Not_Use_When>
17	- User wants to find academic papers specifically (use /paper-discover)
18	- User wants to process an existing vault note (use /process)
19	- User wants to synthesize across existing vault notes (use /synthesize)
20	- User has a YouTube video to process (use /youtube)
21	</Do_Not_Use_When>
22
23	<Steps>
24
25	## Stage 1: PLAN — Decompose the Question
26
27	Parse the topic from $ARGUMENTS. If vague, ask one clarifying question.
28
29	Break the research topic into 3-5 specific sub-questions that together
30	cover the topic well. Think about:
31	- What IS it? (definition, core mechanism)
32	- Why does it matter? (motivation, impact, who cares)
33	- How does it work? (process, architecture, method)
34	- What are the tradeoffs? (limitations, alternatives, open problems)
35	- Where is it going? (trends, recent developments, future)
36
37	Present the sub-questions to the user briefly:
38	```
39	Researching "topic". Sub-questions:
40	1. ...
41	2. ...
42	3. ...
43	Searching now.
44	```
45
46	Don't wait for confirmation unless the topic is ambiguous — just show and go.
47
48	## Stage 2: SEARCH VAULT — What Do We Already Know?
49
50	Before hitting the web, check what the vault already contains:
51
52	```
53	Grep(pattern="KEYWORD", path="notes/", glob="*.md", head_limit=15)
54	```
55
56	Also try MCP search if available:
57	```
58	mcp__obsidian-vault__search_notes(query="KEYWORD", limit=10)
59	```
60
61	Note any existing vault notes that are relevant — these become [[wikilinks]]
62	in the output and inform what the web search should FOCUS on (gaps, not
63	repeats).
64
65	## Stage 3: SEARCH WEB — Multi-Query, Parallel
66
67	For each sub-question, run a targeted WebSearch:
68
69	```
70	WebSearch(query="specific sub-question query", num_results=5)
71	```
72
73	Then WebFetch the 3-5 most promising URLs to get full content:
74
75	```
76	WebFetch(url="URL", prompt="Extract key facts, data, and insights about [sub-question]. Include specific numbers, dates, names, and technical details.")
77	```
78
79	Search strategy:
80	- Use different query phrasings per sub-question (not just the topic repeated)
81	- Prefer recent sources (add "2025" or "2026" to queries when freshness matters)
82	- Mix source types: technical blogs, official docs, news, research summaries
83	- If initial results are thin, reformulate and search again
84
85	Run searches in parallel where possible (multiple WebSearch calls in one turn).
86
87	## Stage 4: DEEPEN — Find Gaps and Conflicts
88
89	After the first search round, review what you have:
90	- Which sub-questions are well-answered? Which are thin?
91	- Are there conflicting claims across sources?
92	- Did any source mention something surprising worth following up?
93
94	Run 1-2 targeted follow-up searches to fill gaps. This second pass is what
95	separates good research from a quick Google.
96
97	## Stage 5: SYNTHESIZE — Build the Note
98
99	Read the agent definition:
100	```
101	Read("${CLAUDE_SKILL_DIR}/agents/research-noter.md")
102	```
103
104	Launch the research-noter agent:
105
106	```
107	Agent(
108	subagent_type="general-purpose",
109	model="sonnet",
110	prompt="You are Research Noter. Follow these instructions exactly:
111
112	[INSERT FULL CONTENT OF agents/research-noter.md HERE]
113
114	RESEARCH TOPIC: [topic]
115	SUB-QUESTIONS: [list]
116
117	VAULT CONTEXT (existing notes on this topic):
118	[vault search results]
119
120	WEB FINDINGS:
121	[organized by sub-question, with source URLs]
122
123	Produce the note body following the Output Format. Do NOT include frontmatter."
124	)
125	```
126
127	## Stage 6: INTEGRATE — Create the Vault Note
128
129	1. Generate timestamp ID: `date +%Y%m%d%H%M%S`
130
131	2. Determine subfolder — if the topic clearly fits an existing folder, use it.
132	Otherwise default to `notes/research/`. Create the folder if needed.
133
134	3. Create the note:
135
136	```markdown
137	---
138	id: YYYYMMDDHHMMSS
139	type: note
140	processing_status: processed
141	created_date: YYYY-MM-DD
142	updated_date: YYYY-MM-DD
143	---
144
145	[AGENT OUTPUT — starts with # title]
146	```
147
148	4. Report to user:
149	- Note path
150	- Number of sources used
151	- Key vault connecti

People who liked this also viewed

6 picks

Type

research

32stars2forks

View on GitHub

How to install research?

tuan3w/obsidian-vault-agent/research

$npx -y skills add tuan3w/obsidian-vault-agent --skill research

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Deep research on any topic — search the web and the vault, then synthesize
3	findings into a high-quality vault note. Unlike a quick web search, this skill
4	decomposes the question, searches multiple angles, checks what the vault already
5	knows, and produces a note with source links, confidence levels, and vault
6	connections.
7	</Purpose>
8
9	<Use_When>
10	- User asks to research a topic ("research quantum error correction")
11	- User wants to understand something new ("what is RLHF and why does it matter?")
12	- User wants a research note created in the vault
13	- User uses /research with a topic
14	</Use_When>
15
16	<Do_Not_Use_When>
17	- User wants to find academic papers specifically (use /paper-discover)
18	- User wants to process an existing vault note (use /process)
19	- User wants to synthesize across existing vault notes (use /synthesize)
20	- User has a YouTube video to process (use /youtube)
21	</Do_Not_Use_When>
22
23	<Steps>
24
25	## Stage 1: PLAN — Decompose the Question
26
27	Parse the topic from $ARGUMENTS. If vague, ask one clarifying question.
28
29	Break the research topic into 3-5 specific sub-questions that together
30	cover the topic well. Think about:
31	- What IS it? (definition, core mechanism)
32	- Why does it matter? (motivation, impact, who cares)
33	- How does it work? (process, architecture, method)
34	- What are the tradeoffs? (limitations, alternatives, open problems)
35	- Where is it going? (trends, recent developments, future)
36
37	Present the sub-questions to the user briefly:
38	```
39	Researching "topic". Sub-questions:
40	1. ...
41	2. ...
42	3. ...
43	Searching now.
44	```
45
46	Don't wait for confirmation unless the topic is ambiguous — just show and go.
47
48	## Stage 2: SEARCH VAULT — What Do We Already Know?
49
50	Before hitting the web, check what the vault already contains:
51
52	```
53	Grep(pattern="KEYWORD", path="notes/", glob="*.md", head_limit=15)
54	```
55
56	Also try MCP search if available:
57	```
58	mcp__obsidian-vault__search_notes(query="KEYWORD", limit=10)
59	```
60
61	Note any existing vault notes that are relevant — these become [[wikilinks]]
62	in the output and inform what the web search should FOCUS on (gaps, not
63	repeats).
64
65	## Stage 3: SEARCH WEB — Multi-Query, Parallel
66
67	For each sub-question, run a targeted WebSearch:
68
69	```
70	WebSearch(query="specific sub-question query", num_results=5)
71	```
72
73	Then WebFetch the 3-5 most promising URLs to get full content:
74
75	```
76	WebFetch(url="URL", prompt="Extract key facts, data, and insights about [sub-question]. Include specific numbers, dates, names, and technical details.")
77	```
78
79	Search strategy:
80	- Use different query phrasings per sub-question (not just the topic repeated)
81	- Prefer recent sources (add "2025" or "2026" to queries when freshness matters)
82	- Mix source types: technical blogs, official docs, news, research summaries
83	- If initial results are thin, reformulate and search again
84
85	Run searches in parallel where possible (multiple WebSearch calls in one turn).
86
87	## Stage 4: DEEPEN — Find Gaps and Conflicts
88
89	After the first search round, review what you have:
90	- Which sub-questions are well-answered? Which are thin?
91	- Are there conflicting claims across sources?
92	- Did any source mention something surprising worth following up?
93
94	Run 1-2 targeted follow-up searches to fill gaps. This second pass is what
95	separates good research from a quick Google.
96
97	## Stage 5: SYNTHESIZE — Build the Note
98
99	Read the agent definition:
100	```
101	Read("${CLAUDE_SKILL_DIR}/agents/research-noter.md")
102	```
103
104	Launch the research-noter agent:
105
106	```
107	Agent(
108	subagent_type="general-purpose",
109	model="sonnet",
110	prompt="You are Research Noter. Follow these instructions exactly:
111
112	[INSERT FULL CONTENT OF agents/research-noter.md HERE]
113
114	RESEARCH TOPIC: [topic]
115	SUB-QUESTIONS: [list]
116
117	VAULT CONTEXT (existing notes on this topic):
118	[vault search results]
119
120	WEB FINDINGS:
121	[organized by sub-question, with source URLs]
122
123	Produce the note body following the Output Format. Do NOT include frontmatter."
124	)
125	```
126
127	## Stage 6: INTEGRATE — Create the Vault Note
128
129	1. Generate timestamp ID: `date +%Y%m%d%H%M%S`
130
131	2. Determine subfolder — if the topic clearly fits an existing folder, use it.
132	Otherwise default to `notes/research/`. Create the folder if needed.
133
134	3. Create the note:
135
136	```markdown
137	---
138	id: YYYYMMDDHHMMSS
139	type: note
140	processing_status: processed
141	created_date: YYYY-MM-DD
142	updated_date: YYYY-MM-DD
143	---
144
145	[AGENT OUTPUT — starts with # title]
146	```
147
148	4. Report to user:
149	- Note path
150	- Number of sources used
151	- Key vault connecti

People who liked this also viewed

6 picks

Type