deep-research

Name: deep-research
Author: tuan3w

32stars2forks

How to install deep-research?

tuan3w/obsidian-vault-agent/deep-research

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

Installs into the current project.

Summary

Deep, iterative, multi-source research that explores a topic from multiple perspectives, detects contradictions, and produces a comprehensive vault note with fact-level citations and a confidence map. Use this skill whenever the user says /deep-research, "deep research X", "resea

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Iterative, multi-perspective research that explores broadly and goes deep. Unlike
3	`/research` (single-pass, web-only, 5-10 sources), this skill:
4	- Discovers PERSPECTIVES before searching (not just sub-questions)
5	- Searches 4 source types in parallel: web, academic papers, Reddit, vault
6	- Runs an exploration → critique loop that finds gaps, contradictions, and new angles
7	- Uses an outline-first approach to prevent hallucination
8	- Produces a note with fact-level citations, a confidence map, and explicit uncertainty
9
10	The architecture is inspired by Stanford's STORM (perspective discovery + outline-first),
11	Anthropic's multi-agent research system (parallel exploration + intelligent critique),
12	and pi-autoresearch (living state files that survive context resets).
13	</Purpose>
14
15	<Use_When>
16	- User wants DEEP understanding, not a quick summary
17	- Topic is complex, contested, or multi-faceted
18	- User needs to make a decision based on the research
19	- User wants to compare competing approaches with evidence
20	- User explicitly asks for /deep-research or "thorough research"
21	</Use_When>
22
23	<Do_Not_Use_When>
24	- User wants a quick answer (use /research instead)
25	- User wants to find academic papers specifically (use /paper-discover)
26	- User wants to process an existing vault note (use /process)
27	- User wants cross-domain vault synthesis (use /synthesize)
28	</Do_Not_Use_When>
29
30	<Steps>
31
32	## Stage 0: SETUP
33
34	Parse `$ARGUMENTS` for the topic and optional depth flag:
35	- `--quick`: faster, fewer rounds, sonnet everywhere, no perspective discovery
36	- `--deep`: more rounds, opus for critic and synthesis, outline reviewed by user
37	- Default (no flag): balanced — perspective discovery, 2-4 rounds, opus for synthesis
38
39	Create the workspace:
40	```bash
41	TOPIC_SLUG=$(echo "$TOPIC" \| tr '[:upper:]' '[:lower:]' \| sed 's/[^a-z0-9]/-/g' \| sed 's/--*/-/g' \| cut -c1-30)
42	WORK_DIR="temp/research-${TOPIC_SLUG}"
43	mkdir -p "$WORK_DIR"/{findings,gaps}
44	```
45
46	Resume check: Before creating, check if the workspace already exists:
47	```bash
48	ls "$WORK_DIR/state.md" 2>/dev/null
49	```
50	If `state.md` exists, read it and ask the user: "Found incomplete research on
51	[topic] (round N). Resume or start fresh?" If resuming, read `state.md` to
52	determine which stage to skip to.
53
54	Tell the user what's happening:
55	```
56	Deep researching "{topic}" ({depth mode}).
57	Setting up workspace at {WORK_DIR}...
58	```
59
60	## Stage 1: PLAN — Decompose + Discover Perspectives
61
62	Read the agent definition:
63	```
64	Read(".claude/skills/deep-research/agents/research-planner.md")
65	```
66
67	Launch the planner agent:
68	```
69	Agent(
70	model="sonnet",
71	prompt="You are Research Planner. Follow these instructions exactly:
72
73	[INSERT FULL CONTENT OF agents/research-planner.md HERE]
74
75	TOPIC: {topic}
76	DEPTH_MODE: {quick\|standard\|deep}
77	WORK_DIR: {WORK_DIR}
78
79	Write state.md and ideas.md to the WORK_DIR."
80	)
81	```
82
83	After the planner completes, read `state.md` briefly and tell the user:
84	```
85	Perspectives identified: {list}
86	Sub-questions: {count}
87	Exploration ideas: {count}
88	Starting exploration...
89	```
90
91	## Stage 2: EXPLORE — Parallel Deep Dives
92
93	Read the agent definition and source tiers reference:
94	```
95	Read(".claude/skills/deep-research/agents/research-explorer.md")
96	Read(".claude/skills/deep-research/references/source-tiers.md")
97	```
98
99	Read `ideas.md` and pick the 3-5 highest-priority unexplored ideas.
100
101	For EACH selected idea, launch an explorer agent in the background:
102	```
103	Agent(
104	model="sonnet",
105	run_in_background=true,
106	prompt="You are Research Explorer. Follow these instructions exactly:
107
108	[INSERT FULL CONTENT OF agents/research-explorer.md HERE]
109
110	SOURCE TIERS REFEREN

People who liked this also viewed

6 picks

Type

deep-research

32stars2forks

View on GitHub

How to install deep-research?

tuan3w/obsidian-vault-agent/deep-research

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Iterative, multi-perspective research that explores broadly and goes deep. Unlike
3	`/research` (single-pass, web-only, 5-10 sources), this skill:
4	- Discovers PERSPECTIVES before searching (not just sub-questions)
5	- Searches 4 source types in parallel: web, academic papers, Reddit, vault
6	- Runs an exploration → critique loop that finds gaps, contradictions, and new angles
7	- Uses an outline-first approach to prevent hallucination
8	- Produces a note with fact-level citations, a confidence map, and explicit uncertainty
9
10	The architecture is inspired by Stanford's STORM (perspective discovery + outline-first),
11	Anthropic's multi-agent research system (parallel exploration + intelligent critique),
12	and pi-autoresearch (living state files that survive context resets).
13	</Purpose>
14
15	<Use_When>
16	- User wants DEEP understanding, not a quick summary
17	- Topic is complex, contested, or multi-faceted
18	- User needs to make a decision based on the research
19	- User wants to compare competing approaches with evidence
20	- User explicitly asks for /deep-research or "thorough research"
21	</Use_When>
22
23	<Do_Not_Use_When>
24	- User wants a quick answer (use /research instead)
25	- User wants to find academic papers specifically (use /paper-discover)
26	- User wants to process an existing vault note (use /process)
27	- User wants cross-domain vault synthesis (use /synthesize)
28	</Do_Not_Use_When>
29
30	<Steps>
31
32	## Stage 0: SETUP
33
34	Parse `$ARGUMENTS` for the topic and optional depth flag:
35	- `--quick`: faster, fewer rounds, sonnet everywhere, no perspective discovery
36	- `--deep`: more rounds, opus for critic and synthesis, outline reviewed by user
37	- Default (no flag): balanced — perspective discovery, 2-4 rounds, opus for synthesis
38
39	Create the workspace:
40	```bash
41	TOPIC_SLUG=$(echo "$TOPIC" \| tr '[:upper:]' '[:lower:]' \| sed 's/[^a-z0-9]/-/g' \| sed 's/--*/-/g' \| cut -c1-30)
42	WORK_DIR="temp/research-${TOPIC_SLUG}"
43	mkdir -p "$WORK_DIR"/{findings,gaps}
44	```
45
46	Resume check: Before creating, check if the workspace already exists:
47	```bash
48	ls "$WORK_DIR/state.md" 2>/dev/null
49	```
50	If `state.md` exists, read it and ask the user: "Found incomplete research on
51	[topic] (round N). Resume or start fresh?" If resuming, read `state.md` to
52	determine which stage to skip to.
53
54	Tell the user what's happening:
55	```
56	Deep researching "{topic}" ({depth mode}).
57	Setting up workspace at {WORK_DIR}...
58	```
59
60	## Stage 1: PLAN — Decompose + Discover Perspectives
61
62	Read the agent definition:
63	```
64	Read(".claude/skills/deep-research/agents/research-planner.md")
65	```
66
67	Launch the planner agent:
68	```
69	Agent(
70	model="sonnet",
71	prompt="You are Research Planner. Follow these instructions exactly:
72
73	[INSERT FULL CONTENT OF agents/research-planner.md HERE]
74
75	TOPIC: {topic}
76	DEPTH_MODE: {quick\|standard\|deep}
77	WORK_DIR: {WORK_DIR}
78
79	Write state.md and ideas.md to the WORK_DIR."
80	)
81	```
82
83	After the planner completes, read `state.md` briefly and tell the user:
84	```
85	Perspectives identified: {list}
86	Sub-questions: {count}
87	Exploration ideas: {count}
88	Starting exploration...
89	```
90
91	## Stage 2: EXPLORE — Parallel Deep Dives
92
93	Read the agent definition and source tiers reference:
94	```
95	Read(".claude/skills/deep-research/agents/research-explorer.md")
96	Read(".claude/skills/deep-research/references/source-tiers.md")
97	```
98
99	Read `ideas.md` and pick the 3-5 highest-priority unexplored ideas.
100
101	For EACH selected idea, launch an explorer agent in the background:
102	```
103	Agent(
104	model="sonnet",
105	run_in_background=true,
106	prompt="You are Research Explorer. Follow these instructions exactly:
107
108	[INSERT FULL CONTENT OF agents/research-explorer.md HERE]
109
110	SOURCE TIERS REFEREN

People who liked this also viewed

6 picks

Type