/extract

Extract Action Items from Text

Extract structured action items from pasted text (meeting transcripts, email threads, or documents) and create Notion tasks automatically.

Load Skill

Read the action-extraction skill at ${CLAUDE_PLUGIN_ROOT}/skills/actions/action-extraction/SKILL.md for verb patterns, owner inference rules, deadline parsing, priority scoring, duplicate detection, and Notion task creation procedures.

Business Context (Optional)

Check if context files exist at _infrastructure/context/active/. If the directory contains .md files, read business-info.md, strategy.md, and current-data.md. Use this context to personalize output (e.g., prioritize known clients, use correct terminology, align with current strategy). If files don't exist, skip silently.

Step 0: Memory Context

Read the context-injection skill at _infrastructure/memory/context-injection/SKILL.md. Query for memories relevant to the current input (company, contacts, topics detected in arguments). If memories are returned, incorporate them into your working context for this execution.

Parse Input

The user's input is provided in $ARGUMENTS. This is the raw text to extract action items from.

If $ARGUMENTS is empty or missing:

• Halt with: "No text provided. Paste the content you want to extract action items from, or use /founder-os:actions:extract-file [path] to process a file."

Extraction Process

1. Detect source type: Analyze the text to determine if it is a meeting transcript, email thread, document, or other format. Use the source type detection rules from the skill.

2. Extract source title: Identify the title or subject of the source content:

   • Meeting transcript: Look for a title line, "Meeting:" header, or subject in metadata
   • Email: Extract the Subject: header
   • Document: Use the first heading or first meaningful line
   • If no title found, use "Untitled [Source Type]"

3. Scan for action items: Apply the three-pass verb pattern scan from the skill (direct requests, implicit actions, commitment language). For meeting transcripts, also apply transcript-specific patterns.

4. Build action items: For each detected action, construct the full action item structure per the skill: title (verb-first, max 80 chars), description, owner (using inference rules), deadline (using parsing rules), priority (1-5), source_type, source_title, context.

5. Deduplicate within batch: Ensure no two action items from the same source have identical titles. Disambiguate by adding specificity from the text.

Notion Integration

1. Database discovery: Search Notion for a database named "[FOS] Tasks". If not found, try "Founder OS HQ - Tasks". If not found, fall back to "Action Item Extractor - Tasks". If none exists, use graceful degradation (below). Do NOT create the database.
2. Duplicate detection: For each action item, query the database for tasks with similar titles created within the past 14 days, filtered by Type = "Action Item". Apply the duplicate detection rules from the skill (verb+noun matching). Skip duplicates, cross-reference near-duplicates.
3. Create tasks: Create a Notion page for each new action item with all properties populated:
   • Type = "Action Item"
   • Source Plugin = "Action Extractor"
   • Title, Description, Owner, Deadline, Priority, Status ("To Do"), Source Type, Source Title, Extracted At (now)
   • Company / Contact relations when owner matches a CRM contact (see skill for matching rules)

Graceful Degradation

If Notion MCP is unavailable or any Notion operation fails:

• Output all extracted action items as structured text in chat
• Format each item clearly with all fields (title, description, owner, deadline, priority, source type, source title)
• Warn: "Notion unavailable — displaying results in chat. Tasks were not saved to Notion. Configure Notion MCP per ${CLAUDE_PLUGIN_ROOT}/INSTALL.md."

Output Format

After processing, display results:

## Action Items Extracted

**Source**: [source_title] ([source_type])
**Items found**: [count]
**Created in Notion**: [count] new | [count] duplicates skipped

---

| # | Action Item | Owner | Deadline | Priority | Status |
|---|-------------|-------|----------|----------|--------|
| 1 | [title] | [owner] | [deadline or "—"] | [priority]/5 | [Created ✓ / Duplicate ⚠️] |
| 2 | ... | ... | ... | ... | ... |

[For each created task, include the Notion page link]

If no action items were found:

• Display: "No action items detected in the provided text. The text may not contain actionable requests, tasks, or commitments."

Final: Memory Update

Read the pattern-detection skill at _infrastructure/memory/pattern-detection/SKILL.md. Log this execution as an observation with: plugin name, primary action performed, key entities (companies, contacts), and output summary. Check for emerging patterns per the detection rules. If a memory reaches the adaptation threshold, append the notification to the output.

Usage Examples

/founder-os:actions:extract Meeting notes from standup: John will review the API spec by Friday. Sarah to update the deployment docs. AI: Schedule follow-up meeting for next week.

/founder-os:actions:extract From: alice@example.com Subject: Q4 Planning Hi team, please review the attached budget by EOD Thursday. Bob, can you prepare the slide deck for Monday's presentation?

/founder-os:actions:extract Next steps from project kickoff: 1. Draft project charter - Mike, by Jan 15 2. Set up development environment - Dev team, this week 3. Schedule stakeholder interviews - PM, ASAP