Inbox learner

Inbox Learning Processor

Instructions

This skill delegates URL processing to the inbox-processor subagent, which runs autonomously with restricted permissions and uses the Haiku model for cost efficiency.

Your task: Launch the inbox-processor agent to handle the work.

Delegation Strategy

Use the Task tool to launch the inbox-processor subagent:

Task:
  subagent_type: inbox-processor
  prompt: "Process all URLs from the Obsidian inbox and create learning summaries. Use today's actual date (YYYY-MM-DD format) for all file names, frontmatter dates, and archive entries."
  description: "Process inbox URLs"
  model: haiku

The agent will autonomously:

1. Read the inbox
2. Fetch and process each URL
3. Create learning entries
4. Link related topics
5. Archive processed URLs
6. Report back with results

Fetching Strategy

Primary: WebFetch

• Fast and lightweight
• Works for most static and server-rendered pages
• No browser overhead

Fallback: Puppeteer Script

• For JavaScript-rendered pages (React, Vue, Framer, etc.)
• Runs in sandboxed Chrome with security enabled
• Located at ${CLAUDE_SKILL_ROOT}/fetch-page.js
• Usage: cd ${CLAUDE_SKILL_ROOT} && node fetch-page.js "https://example.com"

Detection: If WebFetch returns mostly CSS/JS or very little text content, automatically try Puppeteer fallback.

Summary Requirements

Keep summaries concise - the goal is to provide enough information to:

• Understand what the content is about
• Decide if you want to read the full article
• Find it later when searching for related topics
• Recall technical details without re-reading the full article

Summary Structure:

• Title: Descriptive title based on content
• Source: Original URL
• Date: Processing date
• Topics: Tags for easy searching (e.g., #security, #prompt-injection, #supply-chain)
• Summary: 2-3 paragraphs capturing the essence
• Key Points: 3-5 bullet points with main takeaways
• Technical Insights: Implementation details, code patterns, techniques, or technical concepts
• Related: Links to other learning entries with similar topics

File Naming Convention

Use descriptive, slug-style names based on the main topic:

• Format: YYYY-MM-DD-topic-name.md where YYYY-MM-DD is today's date
• Example (if today is 2026-03-03): 2026-03-03-prompt-injection-cline.md
• Example (if today is 2026-03-03): 2026-03-03-github-repo-data-leak.md

IMPORTANT: Always use today's actual date, not the example dates shown above.

Topic Extraction

Extract topics that help with discoverability:

• Technology: Programming languages, frameworks, tools
• Concepts: Design patterns, security concepts, best practices
• Domain: Security, DevOps, AI, databases, etc.
• Specific: Product names, vulnerabilities, techniques

Linking Strategy

After creating each entry:

1. Search existing Learning/ entries for related topics using mcp__mcp-obsidian-myjournal__obsidian_simple_search
2. If related entries found (matching topics/tags):
   • Add link to new entry: Related: [[other-entry]]
   • Update related entry to link back: See also: [[new-entry]]
3. This creates a knowledge graph for easy navigation

Archive Process

After successfully processing each URL:

1. Read Archive: Get contents of 00_Inbox_Archive.md (create if doesn't exist)
2. Append Entry: Add to archive under the current month heading with format:

   - [YYYY-MM-DD] [URL] → [[learning-entry-name]]
   

3. Remove from Inbox: Delete the URL line from 00_Inbox.md
   • Note: Obsidian MCP tools don't support file overwrite
   • Workaround: Read current inbox, filter out processed URLs, then use obsidian_delete_file to delete the old file and create a new one with cleaned content
   • This ensures the inbox stays clean after processing

IMPORTANT - Monthly Grouping:

• Entries are grouped by month name (e.g., ### March), NOT by day
• Use heading format: ### Month Name (e.g., ### March, ### February)
• Extract the month name from today's date: If today is 2026-03-10, use ### March
• NEVER create day-based headings like ### 2026-03-04
• All entries from the same month go under the same month heading
• The date in brackets [YYYY-MM-DD] provides the specific day information

Archive Structure (example showing entries from different dates in the same month):

# Processed Inbox Items

Archive of URLs processed from the inbox with links to their learning entries.

## 2026

### March
- [2026-03-03] https://example.com/article → [[2026-03-03-topic-name]]
- [2026-03-15] https://example.com/another → [[2026-03-15-another-topic]]
- [2026-03-28] https://example.com/third → [[2026-03-28-third-topic]]

### February
- [2026-02-26] https://example.com/other → [[2026-02-26-other-topic]]

### January
- [2026-01-15] https://example.com/old-article → [[2026-01-15-old-topic]]

Note: Use today's actual date when adding new entries, not the example dates above.

This provides:

• Audit trail: See what was processed and when
• Quick access: Jump to learning entries from URLs
• History: Track learning over time

Template for Knowledge Entries

IMPORTANT: Replace YYYY-MM-DD with today's actual date in all fields below.

---
date: YYYY-MM-DD
source: [original URL]
topics: [tag1, tag2, tag3]
---

# [Title]

**Source**: [Article Title](URL)
**Date Processed**: YYYY-MM-DD
**Topics**: #tag1 #tag2 #tag3

## Summary

[2-3 concise paragraphs explaining what this is about]

## Key Points

- Point 1
- Point 2
- Point 3

## Technical Insights

[Technical details, implementation notes, code patterns, or specific techniques mentioned]

Examples:
- Code snippets or commands
- Configuration patterns
- Attack vectors or defense mechanisms
- API usage examples
- Architectural patterns

## Related

- [[other-entry-name]] - Brief context on the connection

Output to User

After processing, provide:

1. Count of URLs processed
2. List of created files with their main topics
3. Any linking relationships established
4. Archive status confirmation
5. Summary of what was learned (optional)

Error Handling

• If URL fails to fetch, note it in archive with error marker and continue
• If no URLs found in inbox, inform user
• If Learning/ folder doesn't exist, create it first
• If archive file doesn't exist, create it with proper structure

Inbox Cleanup Implementation

Obsidian MCP Limitation: The available Obsidian MCP tools (obsidian_append_content, obsidian_get_file_contents, etc.) don't include a file overwrite capability.

Workaround for cleaning the inbox:

1. Read current 00_Inbox.md content
2. Parse and identify processed URL lines
3. Create new content without processed URLs
4. Delete the old file using obsidian_delete_file
5. Create new file with cleaned content using the Write tool or recreate via append

This ensures the inbox stays clean after each processing run.

Important Notes

• Conciseness is key: Don't replicate the full article, just capture the essence
• Technical depth: Include enough technical detail to jog memory without re-reading
• Searchability: Use consistent topic tags
• Connections: The real value is in linking related concepts
• Audit trail: Archive maintains history of learning journey
• Iterative: The knowledge base grows more valuable over time
• File recreation: Since Obsidian MCP can't overwrite files, delete and recreate when cleaning the inbox

---