Curator Notes Processing

Process the human curator's monthly notes file into pipeline-ready content and editorial signals.

Quick Start

1. •Check for workspace/curator_notes_YYYY-MM.md (or known variants)
2. •If no file exists, skip this phase entirely -- pipeline continues normally
3. •Parse and classify every item using the link type taxonomy
4. •Filter: internal links become editorial signals; public links get visited
5. •Output processed items + editorial signals
6. •Items feed into Phase 3 curation alongside Phase 1C discoveries

Inputs

• •Curator notes file (optional): workspace/curator_notes_YYYY-MM.md
  • •May also appear as: workspace/<month>.md, workspace/<Month>.md, workspace/Jan.md, etc.
  • •Check for any .md file in workspace/ that looks like a brain dump of links
• •Phase 1C Discoveries: For cross-reference (avoid duplicates)

Output

• •workspace/curator_notes_processed_YYYY-MM.md -- Items in universal extraction format, ready for Phase 3
• •workspace/curator_notes_editorial_signals_YYYY-MM.md -- Priority hints, theme suggestions, internal context

Core Workflow

Step 0: Locate the Notes File

Check these paths in order:

1. •workspace/curator_notes_YYYY-MM.md (canonical)
2. •workspace/<Month>.md (e.g., workspace/Jan.md, workspace/February.md)
3. •Any .md file in workspace/ containing 3+ URLs that isn't a pipeline output

If no file found: SKIP. Output nothing. Pipeline continues.

Step 1: Parse and Classify

For each line/item in the notes file, classify into one of these types:

Step 2: Filter and Route

PROCESS items (visit, extract, output as discoveries):

• •GitHub Changelog, Blog, VS Code, DevBlog, Community, Extensions, Videos, Events
• •Visit each URL and extract: title, date, description, enterprise relevance

SIGNAL items (output as editorial signals):

• •Internal links: identify the topic and note "watch for public announcement about [topic]"
• •Industry content: summarize the thesis for framing context
• •Named feature hints: flag as lead section candidates
• •Competitor content: note competitive positioning opportunity

Step 3: Cross-Reference with Phase 1C

For each PROCESS item, check if Theme 1C discoveries already cover it:

• •Already covered: Mark as "REINFORCEMENT" (human agreed this is important -- boost priority in Phase 3)
• •Not covered: Mark as "ADDITION" (unique content from curator -- must include in Phase 3)

Step 4: Visit and Extract

For each PROCESS item not already covered by Phase 1C:

1. •Fetch the URL
2. •Extract: title, date, description (2-3 sentences), enterprise relevance (1-10)
3. •Format in universal extraction format
4. •Assign category from the newsletter taxonomy

Step 5: Write Outputs

Processed items (workspace/curator_notes_processed_YYYY-MM.md):

## Curator Notes: Processed Items

### Additions (not in Phase 1C)
- **[Item Title]** -- Description. - [Label](URL)
  - Source: Curator notes
  - Enterprise Relevance: N/10

### Reinforcements (also in Phase 1C -- boost priority)
- **[Item Title]** -- Already in discoveries. Curator explicitly flagged.

Editorial signals (workspace/curator_notes_editorial_signals_YYYY-MM.md):

## Editorial Signals from Curator Notes

### Lead Section Candidates
- [Feature hint]: Curator named this explicitly. Consider for lead.

### Priority Boosts
- [Topic]: Curator linked N items about this topic. Strong signal.

### Framing Context
- [Industry article summary]: Useful for newsletter introduction framing.

### Internal Signals (not linkable)
- [Internal thread topic]: Something is happening around [X]. Watch for public announcement.

Integration with Pipeline

• •Phase 1C -> Phase 1.5 (this skill) -> Phase 3
• •Phase 3 (content-curation) reads BOTH Phase 1C discoveries AND curator notes processed items
• •Editorial signals inform Phase 3's lead section decision and item prioritization
• •If a named feature hint matches a Phase 1C discovery, it gets 2.0x priority boost

Link Type Survival Intelligence

From analysis of 10 benchmark cycles (see reference/curator-notes-intelligence.md):

Reference

• •Curator Notes Intelligence -- Full analysis from 10 benchmark cycles
• •Content Format Spec -- Output formatting rules

Done When

• • Notes file located (or confirmed absent)
• • All items classified by type
• • Public URLs visited and extracted
• • Cross-referenced with Phase 1C discoveries
• • Processed items file exists (if notes file existed)
• • Editorial signals file exists (if notes file existed)
• • No internal links leaked to processed items output