You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: .opencode/agents/scout.md
+18-31Lines changed: 18 additions & 31 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -2,9 +2,15 @@
2
2
description: Web research subagent - gather external knowledge in parallel
3
3
mode: subagent
4
4
permissions:
5
-
"*": false
5
+
"*": deny
6
6
webfetch: allow
7
7
websearch: allow
8
+
read:
9
+
"*": deny
10
+
.knowledge/notes/*: allow
11
+
edit:
12
+
"*": deny
13
+
.knowledge/notes/*: allow
8
14
---
9
15
10
16
# Scout Subagent
@@ -15,42 +21,23 @@ You are a **Scout** — gathering external knowledge efficiently.
15
21
Research a specific angle of a topic in 60 seconds or less.
16
22
17
23
## Input Format
18
-
```
19
-
Topic: [main research topic]
20
-
Angle: [specific aspect to investigate]
21
-
Context: [why this matters to the project]
22
-
```
24
+
25
+
Read the corresponding notes markdown file to understand the research task in detail.
26
+
27
+
Make sure you understand:
28
+
29
+
-**Topic**: [main research topic]
30
+
-**Angle**: [specific aspect to investigate]
31
+
-**Context**: [why this matters to the project]
23
32
24
33
## Your Workflow
34
+
25
35
1.**Receive assignment** — Understand topic, angle, and context
26
36
2.**Research efficiently** — Web search, docs, authoritative sources
27
37
3.**Extract key insights** — What matters most for this context?
28
38
4.**Return structured** — Compressed intelligence, not raw data
29
-
30
-
## Output Format
31
-
```yaml
32
-
---
33
-
angle: [the angle you researched]
34
-
relevance_score: 0.0-1.0
35
-
sources:
36
-
- [url or reference]
37
-
---
38
-
## Summary
39
-
[1-2 paragraph maximum]
40
-
41
-
## Key Finding
42
-
[Single most important insight for the project, detailed]
43
-
44
-
## Supporting Evidence
45
-
- [point 1]
46
-
- [point 2]
47
-
48
-
## Relevance
49
-
[Why this matters to the specific context]
50
-
```
39
+
5.**Write** to the corresponding notes markdown file with the recommended structure.
51
40
52
41
## Key Mandates
53
-
-**60 second timeout** — Be concise and targeted
54
-
-**No project writes** — Return data to parent only
55
-
-**Compressed output** — Parent will synthesize multiple scouts
56
42
-**Authoritative sources** — Favor official docs, established patterns, academic literature, primary sources.
43
+
-**Depth over Breadth** — You are intended to go deep into one specific search questions. If further questions arise, report them in your final response, but don't follow them. The primary agent will launch aditional searches if necessary.
description: Help using the OpenCode Opinionated Framework
3
+
agent: analyze
4
+
---
5
+
6
+
# OpenCode Opinionated Framework
7
+
8
+
USER REQUEST: $ARGUMENTS
9
+
10
+
**Option 1:** If the command is run without arguments:
11
+
12
+
Explain to the user what this framework is, and how to use it. Do not perform ANY further analysis or read any files, just reply in the following format:
13
+
14
+
```
15
+
## Framework
16
+
17
+
[explain this framework in 1 paragraph]
18
+
19
+
## Main Modes (Agents)
20
+
21
+
[list each primary agent (mode) and its purpose]
22
+
23
+
### Subagents
24
+
25
+
[explain subagents are launched for background tasks automatically, list each one and when it's used]
26
+
27
+
## Commands
28
+
29
+
[list each command and its purpose]
30
+
31
+
## Contextual Analysis
32
+
33
+
[judging from the recent conversation, suggest what the user might want to do, 2-3 commands or instructions to give primary agents]
34
+
```
35
+
36
+
**Option 2:** If the command is run with arguments:
37
+
38
+
Perform 2-3 targeted reads in .opencode subfolder to better understand how some agent or command works.
39
+
40
+
Reply to the user in detail.
41
+
42
+
**Finally:**
43
+
44
+
Finally, ask the user if they have a further question about the framework. If so, then repeat option 2 as necessary.
Create `.knowledge/notes/research-{slug}/` to store temporary notes with boilerplate markdown notes for each research question. Write down in each note the minimal context and relevant subquestions to address.
27
21
28
-
5.**Scout 5: Implementation Examples**
29
-
- Real-world code, tutorials
22
+
### Phase 2: Scout Deployment
30
23
31
-
### Phase 2: Synthesis
32
-
Collect scout findings and synthesize:
33
-
- Merge overlapping insights
34
-
- Resolve contradictions (note conflicts)
35
-
- Identify consensus vs. debate
36
-
- Extract actionable recommendations
24
+
**Perform the following iteration several times until satisfied:**
25
+
26
+
Launch `scout` subagents in parallel for each research question. Point each scout to the corresponding research note, where they will write their output.
27
+
28
+
**Evaluate** for every scout that finishes if further investigation is necessary on any research questions, and repeat until satisfied.
29
+
30
+
If necessary, use `todowrite` to add follow up relevant questions, and iterate.
37
31
38
32
### Phase 3: Documentation
39
-
Write structured research to `.knowledge/insights/research/{slug}.md`:
40
33
41
-
```yaml
34
+
Read all notes files in `.knowledge/notes/{slug}/*.md`.
35
+
36
+
Write structured research to `.knowledge/notes/report-{slug}.md`:
37
+
38
+
```markdown
42
39
---
43
40
id: {kebab-case-topic}
44
41
created: {date}
@@ -56,6 +53,8 @@ sources: [list all scout sources]
56
53
1.**[Finding 1]**
57
54
- Evidence: [from which scout]
58
55
- Implication: [what this means]
56
+
- Link to relevant section in notes markdown.
57
+
- Link to external sources.
59
58
60
59
2.**[Finding 2]**
61
60
...
@@ -64,12 +63,16 @@ sources: [list all scout sources]
64
63
-[Specific, actionable recommendation]
65
64
-[Specific, actionable recommendation]
66
65
67
-
## Scout Reports
68
-
[Summaries from each scout]
66
+
## Further reading
67
+
[Suggest follow up reading on scout-reported sources for adjacent topics]
68
+
69
+
## Follow-up questions
70
+
71
+
[Suggest follow-up research questions that weren't explored.]
0 commit comments