personal memory agent
Simplify category extraction prompts for better OCR focus
Rewrite messaging, browsing, reading, and productivity prompts to be
concise and extraction-focused. Each prompt reduced by 40-57% while
maintaining the essential guidance for text capture and searchability.
- Remove verbose formatting rules the model handles naturally
- Cut detailed sub-sections for edge cases (threads, reactions, etc.)
- Establish consistent structure: goal, header, content focus, quality
- Keep standard markers for unclear/cut-off text
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
+37
-142
+10
-38
observe/categories/browsing.txt
+10
-38
observe/categories/browsing.txt
···
1
1
# Web Browsing Text Extraction
2
2
3
-
You are extracting text from a screenshot of web browsing activity (news sites, blogs, social media feeds, shopping, general websites).
4
-
5
-
Extract the meaningful content and format it as markdown.
3
+
Extract text from this web browsing screenshot (news, blogs, social media, shopping, general websites).
6
4
7
5
## Header
8
6
9
-
Start with the site and page context:
10
-
`# [Site Name - Page Title or Topic]`
11
-
12
-
## Content Prioritization
13
-
14
-
Focus on the primary content, skip:
15
-
- Navigation menus and headers
16
-
- Sidebar widgets and ads
17
-
- Cookie banners and popups
18
-
- Footer links
19
-
20
-
## Article/Post Content
21
-
22
-
For articles and blog posts:
23
-
- Use `##` for article headlines
24
-
- Preserve paragraph structure
25
-
- Include author and date if visible
26
-
- Extract image captions as `*[Image: caption]*`
7
+
`# [Site Name - Page Title]`
27
8
28
-
## Social Media Feeds
9
+
## Content Focus
29
10
30
-
For feed content (Twitter/X, Reddit, Facebook, etc.):
31
-
```markdown
32
-
**@username** (timestamp):
33
-
Post content here
11
+
Extract the primary page content. Skip navigation menus, sidebars, ads, cookie banners, and footers.
34
12
35
-
[12 likes, 3 comments]
36
-
```
13
+
For articles: use `##` for headlines, preserve paragraph structure.
37
14
38
-
## Shopping/Product Pages
15
+
For social feeds: include username, post content, and engagement counts if visible.
39
16
40
-
For e-commerce content:
41
-
- Product name as heading
42
-
- Price prominently noted
43
-
- Key specifications in list format
44
-
- Review snippets if visible
17
+
For product pages: include product name, price, and key details.
45
18
46
-
## Quality Standards
19
+
## Quality
47
20
48
21
- Preserve reading order (main content first)
49
-
- Skip repeated navigation elements
50
22
- Mark unclear text with `[unclear]`
51
-
- Mark cut-off content with `...`
23
+
- Mark cut-off text with `...`
52
24
53
-
Return ONLY the formatted markdown. Do NOT wrap in code fences or JSON.
25
+
Return ONLY the formatted markdown.
+8
-27
observe/categories/messaging.txt
+8
-27
observe/categories/messaging.txt
···
1
1
# Messaging App Text Extraction
2
2
3
-
You are extracting text from a screenshot of a messaging or email application (Slack, Discord, Messages, Gmail, Teams chat, etc.).
4
-
5
-
Extract all visible conversation content and format it as markdown.
3
+
Extract text from this messaging or email screenshot (Slack, Discord, Messages, Gmail, Teams, etc.).
6
4
7
5
## Header
8
6
9
-
Start with the application and channel/conversation context:
10
7
`# [App Name - Channel/Conversation]`
11
8
12
9
## Conversation Format
13
10
14
-
Extract messages with participant attribution:
11
+
Extract messages with sender attribution:
15
12
16
13
```markdown
17
14
**Alice**: Hey, how's it going?
18
15
**Bob**: Pretty good, working on the new feature
19
-
**Alice**: Nice! Let me know if you need help
20
16
```
21
17
22
-
- Use **bold** for participant names
23
-
- Preserve message order (oldest to newest visible)
24
-
- Include timestamps if clearly visible: `**Alice** (2:34 PM): message`
25
-
- Mark reactions/emoji responses inline: `**Bob**: Sounds good [thumbs up from Alice]`
18
+
Include timestamps if visible: `**Alice** (2:34 PM): message`
26
19
27
-
## Thread Context
20
+
Use `>` blockquotes for quoted/forwarded messages. Use code fences for code snippets.
28
21
29
-
If viewing a thread or reply chain:
30
-
- Note the parent message context
31
-
- Indent replies or mark as thread replies
32
-
33
-
## Special Elements
34
-
35
-
- **Quoted/forwarded messages**: Use `>` blockquote
36
-
- **Code snippets**: Use appropriate code fences
37
-
- **Links**: Include the visible link text
38
-
- **Attachments**: Note as `[Attachment: filename or description]`
39
-
- **Unread indicators**: Note where unread messages begin
40
-
41
-
## Quality Standards
22
+
## Quality
42
23
43
24
- Focus on message content, skip UI chrome
44
-
- Preserve conversation flow and context
25
+
- Preserve conversation order and flow
45
26
- Mark unclear text with `[unclear]`
46
-
- Mark cut-off messages with `...`
27
+
- Mark cut-off text with `...`
47
28
48
-
Return ONLY the formatted markdown. Do NOT wrap in code fences or JSON.
29
+
Return ONLY the formatted markdown.
+10
-42
observe/categories/productivity.txt
+10
-42
observe/categories/productivity.txt
···
1
1
# Productivity App Text Extraction
2
2
3
-
You are extracting text from a screenshot of productivity software (spreadsheets, slides, calendars, task managers, issue trackers, project management tools).
4
-
5
-
Extract all visible content with appropriate structure as markdown.
3
+
Extract text from this productivity screenshot (spreadsheets, slides, calendars, task managers, issue trackers, project management tools).
6
4
7
5
## Header
8
6
9
-
Start with the application and context:
10
7
`# [App Name - Document/View Title]`
11
8
12
-
## Spreadsheet Content
13
-
14
-
For spreadsheet views:
15
-
- Use markdown tables for visible data
16
-
- Note column/row headers
17
-
- Include cell formulas if visible: `=SUM(A1:A10)`
18
-
- Note selected cells or ranges
19
-
20
-
```markdown
21
-
| Name | Status | Due Date |
22
-
|------|--------|----------|
23
-
| Task 1 | Done | 2024-01-15 |
24
-
| Task 2 | In Progress | 2024-01-20 |
25
-
```
26
-
27
-
## Presentation Slides
28
-
29
-
For slide content:
30
-
- `##` for slide title
31
-
- Bullet points for slide content
32
-
- Note speaker notes if visible
33
-
- Describe images/charts: `*[Chart: Sales by Quarter]*`
9
+
## Content Focus
34
10
35
-
## Calendar Views
11
+
Extract all visible data with appropriate structure:
36
12
37
-
For calendar content:
38
-
- Note the date range visible
39
-
- List events with times: `**9:00 AM** - Team Standup`
40
-
- Include event details if expanded
41
-
42
-
## Task/Issue Trackers
43
-
44
-
For project management views:
45
-
- Task/issue titles and IDs
46
-
- Status, assignee, priority
47
-
- Due dates and labels
48
-
- Preserve board/list structure
13
+
- **Tables/Spreadsheets**: Use markdown tables, include headers
14
+
- **Calendars**: Include event times and titles (`**9:00 AM** - Team Standup`)
15
+
- **Tasks/Issues**: Include title, status, assignee, and due date if visible
16
+
- **Slides**: Use `##` for slide titles, bullets for content
49
17
50
-
## Quality Standards
18
+
## Quality
51
19
52
20
- Preserve data relationships and hierarchy
53
21
- Include key metadata (dates, statuses, assignees)
54
22
- Mark unclear text with `[unclear]`
55
-
- Mark cut-off content with `...`
23
+
- Mark cut-off text with `...`
56
24
57
-
Return ONLY the formatted markdown. Do NOT wrap in code fences or JSON.
25
+
Return ONLY the formatted markdown.
+9
-35
observe/categories/reading.txt
+9
-35
observe/categories/reading.txt
···
1
1
# Document Reading Text Extraction
2
2
3
-
You are extracting text from a screenshot of document reading activity (PDFs, documentation, articles, ebooks, manuals).
4
-
5
-
Extract all visible text with proper document structure as markdown.
3
+
Extract text from this document screenshot (PDFs, documentation, articles, ebooks, manuals).
6
4
7
5
## Header
8
6
9
-
Start with the document context:
10
7
`# [Document Title or Type]`
11
8
12
-
## Document Structure
13
-
14
-
Preserve the document hierarchy:
15
-
- Use `##`, `###` for section headings
16
-
- Maintain numbered sections (e.g., "2.3 Configuration")
17
-
- Preserve list formatting (bullets, numbered lists)
18
-
- Keep table structure using markdown tables
19
-
20
-
## Code and Technical Content
21
-
22
-
For documentation with code:
23
-
```language
24
-
code here
25
-
```
26
-
27
-
For command examples:
28
-
```bash
29
-
$ command example
30
-
```
31
-
32
-
## References and Citations
9
+
## Content Focus
33
10
34
-
- Include footnotes as `[1]` with reference text if visible
35
-
- Preserve figure/table captions: `*Figure 3: Description*`
36
-
- Note page numbers if visible: `[Page 42]`
37
-
38
-
## Academic/Formal Documents
11
+
Extract all visible text, preserving document structure:
39
12
40
-
- Preserve abstract/summary sections
41
-
- Include author/date information
42
-
- Note section numbers
13
+
- Use `##`, `###` for section headings
14
+
- Preserve numbered sections (e.g., "2.3 Configuration")
15
+
- Use markdown tables for tabular data
16
+
- Use code fences with language tags for code examples
43
17
44
-
## Quality Standards
18
+
## Quality
45
19
46
20
- Maintain logical reading flow
47
21
- Preserve paragraph breaks
···
49
23
- Mark unclear text with `[unclear]`
50
24
- Mark cut-off text with `...`
51
25
52
-
Return ONLY the formatted markdown. Do NOT wrap in code fences or JSON.
26
+
Return ONLY the formatted markdown.