mirror of
https://github.com/rowboatlabs/rowboat.git
synced 2026-04-29 18:36:23 +02:00
add chrome_sync as source for knowledge graph
- Add chrome_sync to SOURCE_FOLDERS in build_graph so captured web pages are picked up alongside emails, transcripts, and notes - Update all three note creation agents (high/medium/low) with browsing source type rules: never create new notes, only update existing ones via a "## From web browsing" section - Browsing entries include author/source, brief description, and URL Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Arjun
parent
de189f1311
commit
a04ea1b92c
2 changed files with 79 additions and 4 deletions
|
|
@ -30,6 +30,7 @@ const SOURCE_FOLDERS = [
|
|||
'gmail_sync',
|
||||
path.join('knowledge', 'Meetings', 'fireflies'),
|
||||
path.join('knowledge', 'Meetings', 'granola'),
|
||||
'chrome_sync',
|
||||
];
|
||||
|
||||
// Voice memos are now created directly in knowledge/Voice Memos/<date>/
|
||||
|
|
@ -182,7 +183,23 @@ async function createNotesFromBatch(
|
|||
message += `- Create or update notes in "knowledge" directory (workspace-relative paths like "knowledge/People/Name.md")\n`;
|
||||
message += `- If the same entity appears in multiple files, merge the information into a single note\n`;
|
||||
message += `- Use workspace tools to read existing notes (when you need full content) and write updates\n`;
|
||||
message += `- Follow the note templates and guidelines in your instructions\n\n`;
|
||||
message += `- Follow the note templates and guidelines in your instructions\n`;
|
||||
|
||||
// Add browsing-specific instructions if any files are from chrome_sync
|
||||
const hasBrowsingFiles = files.some(f => f.path.includes('chrome_sync'));
|
||||
if (hasBrowsingFiles) {
|
||||
message += `\n**BROWSING SOURCE RULES (chrome_sync files):**\n`;
|
||||
message += `- Files from chrome_sync are captured web pages from the user's browser\n`;
|
||||
message += `- They have frontmatter with url, title, and captured_at fields\n`;
|
||||
message += `- NEVER create new notes from browsing data — only update EXISTING notes\n`;
|
||||
message += `- Read the page content and determine which existing People, Organizations, Projects, or Topics are relevant\n`;
|
||||
message += `- For each relevant existing note, add an entry under a "## From web browsing" section\n`;
|
||||
message += `- Browsing entry format: \`- **{YYYY-MM-DD}**: {Author/source} — {Brief description} — {URL}\`\n`;
|
||||
message += `- Always identify who posted/authored the content when possible (author names, bylines, LinkedIn profile names, etc.)\n`;
|
||||
message += `- Add the ## From web browsing section after ## Activity if it doesn't exist yet\n`;
|
||||
message += `- If the browsed content is not relevant to any existing note, skip it\n`;
|
||||
}
|
||||
message += `\n`;
|
||||
|
||||
// Add the knowledge base index
|
||||
message += `---\n\n`;
|
||||
|
|
|
|||
|
|
@ -178,10 +178,15 @@ workspace-readFile({ path: "{source_file}" })
|
|||
- Has frontmatter \`path:\` field like \`Voice Memos/YYYY-MM-DD/...\`
|
||||
- Has \`## Transcript\` section
|
||||
|
||||
**Browsing indicators (chrome_sync):**
|
||||
- Has frontmatter with \`url:\`, \`title:\`, and \`captured_at:\` fields
|
||||
- File path contains \`chrome_sync/\`
|
||||
|
||||
**Set processing mode:**
|
||||
- \`source_type = "meeting"\` → Can create new notes
|
||||
- \`source_type = "email"\` → Can create notes if personalized and relevant
|
||||
- \`source_type = "voice_memo"\` → Can create new notes (treat like meetings)
|
||||
- \`source_type = "browsing"\` → Can ONLY update existing notes, adds to ## From web browsing section
|
||||
|
||||
---
|
||||
|
||||
|
|
@ -249,6 +254,26 @@ labeled_at: "2026-02-28T12:00:00Z"
|
|||
|
||||
${renderNoteEffectRules()}
|
||||
|
||||
## From web browsing Source Filtering (chrome_sync)
|
||||
|
||||
For browsing sources, the rules are simple:
|
||||
- **NEVER create new notes** from browsing data
|
||||
- Read the captured page content, title, and URL
|
||||
- Determine which existing notes (People, Organizations, Projects, Topics) the page is relevant to
|
||||
- If no existing notes are relevant → SKIP
|
||||
- If relevant notes exist → Continue processing, will add to ## From web browsing section only
|
||||
|
||||
**What makes a page relevant to a note?**
|
||||
- Page is about a person who has an existing note (e.g., their LinkedIn profile, blog post)
|
||||
- Page is about an organization that has an existing note (e.g., company website, news article)
|
||||
- Page is about a project or topic that has an existing note
|
||||
- Page contains information directly related to an entity in your knowledge base
|
||||
|
||||
**Skip browsing pages that are:**
|
||||
- Generic content not related to any existing entity
|
||||
- Your own company's pages
|
||||
- General news, tutorials, or documentation not tied to a specific entity
|
||||
|
||||
## Filter Decision Output
|
||||
|
||||
If skipping:
|
||||
|
|
@ -852,15 +877,47 @@ workspace-edit({
|
|||
})
|
||||
\`\`\`
|
||||
|
||||
## 9b: Apply State Changes
|
||||
## 9b: Browsing — Update Existing Notes Only (## From web browsing Section)
|
||||
|
||||
**Only update notes that already exist. NEVER create new notes from browsing data.**
|
||||
|
||||
For each existing note that is relevant to the browsed page:
|
||||
1. Read the current note
|
||||
2. If no \`## From web browsing\` section exists, add one after \`## Activity\`
|
||||
3. Add a new entry at the TOP of the Browsing section (reverse chronological)
|
||||
|
||||
**Entry format:**
|
||||
\`\`\`markdown
|
||||
## From web browsing
|
||||
- **{YYYY-MM-DD}**: {Author/source if identifiable} — {Brief 1-sentence description of the page content} — {URL}
|
||||
\`\`\`
|
||||
|
||||
**Example:**
|
||||
\`\`\`markdown
|
||||
## From web browsing
|
||||
- **2025-01-20**: Acme Corp blog — Announced Series B funding round of $50M — https://acme.com/blog/series-b
|
||||
- **2025-01-15**: Sarah Chen (LinkedIn) — Profile showing recent promotion to VP Engineering — https://linkedin.com/in/sarahchen
|
||||
- **2025-01-14**: John Park (LinkedIn post) — Shared thoughts on MCP as backbone of enterprise AI architecture — https://linkedin.com/feed/
|
||||
\`\`\`
|
||||
|
||||
**Guidelines:**
|
||||
- Always include who posted/authored the content when identifiable from the page text (look for author names, "Posted by", bylines, LinkedIn profile names, etc.)
|
||||
- Keep descriptions brief (one sentence)
|
||||
- Always include the URL if available from the frontmatter
|
||||
- Use the \`captured_at\` date from frontmatter for the date
|
||||
- Do NOT update Activity, Key facts, Open items, or other sections from browsing data
|
||||
- Only update the ## From web browsing section
|
||||
- Update \`**Last seen:**\` date in Info section
|
||||
|
||||
## 9c: Apply State Changes
|
||||
|
||||
For each state change identified in Step 7, update the relevant fields.
|
||||
|
||||
## 9c: Update Aliases
|
||||
## 9d: Update Aliases
|
||||
|
||||
If you discovered new name variants during resolution, add them to Aliases field.
|
||||
|
||||
## 9d: Writing Rules
|
||||
## 9e: Writing Rules
|
||||
|
||||
- **Always use absolute paths** with format \`[[Folder/Name]]\` for all links
|
||||
- Use YYYY-MM-DD format for dates
|
||||
|
|
@ -909,6 +966,7 @@ ${renderNoteTypesBlock()}
|
|||
| Voice memo | Yes | Yes | Yes |
|
||||
| Email (has create label) | Yes | Yes | Yes |
|
||||
| Email (only skip labels) | No (SKIP) | No | No |
|
||||
| Browsing (chrome_sync) | No | Yes (## From web browsing only) | No |
|
||||
|
||||
**Meeting activity format:** Always include a link to the source meeting note:
|
||||
\`\`\`
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue