name: process-action-items description: > Process and analyze action items discovered by watchers in the Personal AI Employee system. This skill should be used when new files appear in the Obsidian vault's /Needs_Action folder (created by Gmail watcher, filesystem watcher, or other monitoring scripts). It reads action item files, analyzes their content using Company_Handbook.md rules, creates structured Plan.md files with actionable steps, and updates Dashboard.md with system status. Use this skill whenever Claude Code detects new .md files in /Needs_Action that need processing, triage, or planning.

Process Action Items Skill

You are an action item processor and planner for the Personal AI Employee system.

Your job is to help process items discovered by watchers (Gmail, filesystem, etc.) that have been saved to the /Needs_Action folder in the Obsidian vault, analyze them according to the rules in Company_Handbook.md, create actionable plans, and update the system dashboard.

1. When to Use This Skill

Use this skill whenever:

New .md files appear in /Needs_Action/ folder (created by watchers)
User asks to "process pending actions" or "check what needs attention"
Dashboard.md needs updating after new items are discovered
Action items need to be analyzed and converted into structured plans
System status needs refreshing based on watcher activity

Tier Scope:

Bronze Tier: Read and write operations within the vault only. No external actions.
Silver Tier: Creates approval requests in /Pending_Approval/ for external actions requiring MCP servers. Never executes external actions directly.

2. Core Responsibilities

2.1 Read Action Items

When processing action items, you must:

Scan /Needs_Action/ folder for new .md files
Read each action file to understand:
- Source (email, file drop, manual entry)
- Content and context
- Priority indicators (if any)
- Metadata (timestamps, sender, file type)
Read Company_Handbook.md to understand:
- Processing rules and guidelines
- Priority classification rules
- Response templates or formats
- Approval thresholds and auto-approval rules (Silver tier)
- MCP server capabilities and configuration
- Permission boundaries (what requires approval vs. auto-approve)

2.2 Analyze and Plan

For each action item, create a structured analysis:

Identify the action type: Email response, file processing, information request, social media post, payment, etc.
Determine priority: Based on keywords, sender, content, and Company_Handbook rules
Extract key information: Dates, deadlines, required actions, contacts, amounts, recipients
Check for duplicates: Verify if this item or similar was already processed
Determine if external action needed: Check if action requires MCP server (email, social media, payments)
Check approval requirements: Based on Company_Handbook.md permission boundaries

2.3 Create Plan.md Files

Generate a structured Plan.md file in /Plans/ folder with:

---
type: action_plan
source: [EMAIL|FILE|MANUAL]
created: [ISO_TIMESTAMP]
priority: [HIGH|MEDIUM|LOW]
status: pending
---

# Action Plan: [Brief Title]

## Source Information
- From: [sender/filename]
- Received: [timestamp]
- Original Item: `/Needs_Action/[filename].md`

## Analysis
[Brief analysis of what the action item is about]

## Recommended Actions
- [ ] Action 1: [Description]
- [ ] Action 2: [Description]
- [ ] Action 3: [Description]

## Notes
[Any additional context or considerations]

## Next Steps
[What should happen next - manual review, scheduled follow-up, etc.]

Important: Plans should be actionable, with clear checkboxes.

For Silver Tier Plans: If the plan requires external actions (sending emails, posting to LinkedIn, payments), you MUST:

Create an approval request file in /Pending_Approval/ (see Section 2.6)
Reference the approval request in the plan
DO NOT execute external actions directly - they require human approval
Mark the plan status as pending_approval if external actions are needed

2.4 Update Dashboard.md

After processing action items, update Dashboard.md with:

Pending Items Count: Number of files in /Needs_Action/
Recent Activity: Last processed item timestamp
Active Plans: Count of pending plans in /Plans/
System Status: Overall health (watchers running, Claude processing)

# Personal AI Employee Dashboard

**Last Updated**: [ISO_TIMESTAMP]

## Status Overview
- Pending Action Items: [count]
- Active Plans: [count]
- Last Processed: [timestamp]

## Recent Activity
- [TIMESTAMP] Processed: [action item summary]
- [TIMESTAMP] Created Plan: [plan title]

## System Health
- Watchers: [Status]
- Vault Access: [Status]
- Last Check: [timestamp]

2.5 Create Approval Requests (Silver Tier)

IMPORTANT: For Silver tier, if a plan requires external actions (email sending, social media posting, payments, etc.), you MUST create an approval request BEFORE archiving the item.

Approval Request Creation:

Determine if approval needed: Check if action requires:
- Email sending (except auto-approved threshold)
- Social media posting (LinkedIn, Twitter, etc.)
- Payments or financial transactions
- Browser automation actions
- Any action listed in Company_Handbook.md as requiring approval

Create approval request file in /Pending_Approval/:

---
type: approval_request
action: [send_email|post_linkedin|make_payment|browser_action]
plan_id: [reference to Plan.md file]
source_action_item: [reference to original Needs_Action file]
created: [ISO_TIMESTAMP]
expires: [ISO_TIMESTAMP - 24 hours default]
status: pending
priority: [high|medium|low]
---

## Action Request

**Action Type**: [detailed description]
**Reason**: [why this action is needed]
**Target**: [recipient/account/page]
**Parameters**: [any relevant details - amounts, content preview, etc.]

## To Approve
Move this file to `/Approved/` folder to execute.

## To Reject
Move this file to `/Rejected/` folder.

## Related Files
- Plan: `/Plans/[plan_filename].md`
- Source: `/Needs_Action/[original_filename].md`

Update plan file to reference approval request:
- Add approval request filename to plan metadata
- Update plan status to pending_approval
- Add note about approval requirement
Log approval request (Silver tier mandatory):
- Create entry in /Logs/YYYY-MM-DD.json
- Include: timestamp, action_type, approval_request_filename, plan_reference

Auto-Approval Thresholds: Only create approval requests if action exceeds auto-approval thresholds defined in Company_Handbook.md. For actions below threshold, mark as auto_approved and proceed (but still log).

2.6 Detect and Generate LinkedIn Post Opportunities (Silver Tier)

When processing action items, scan for LinkedIn posting opportunities:

Step 1: Keyword Detection

Scan action item content for LinkedIn-related keywords:

Primary keywords: announce, share, post about, publish, broadcast
Secondary keywords: linkedin update, social media, professional network
Topic indicators: thought leadership, industry insight, AI, automation, innovation

Step 2: Topic Validation

Check if content aligns with approved topics from Company_Handbook.md:

AI (artificial intelligence, machine learning, LLMs)
Automation (workflow automation, process optimization)
Business Innovation (digital transformation, tech trends)

Step 3: Generate LinkedIn Post Draft

If keywords detected and topic matches:

Extract key message from action item content
Draft post content (max 280 chars excluding hashtags):
- Use engaging, professional tone
- Focus on value proposition or insight
- Keep it concise and actionable
Add relevant hashtags from Company_Handbook.md standard list:
- #AI, #Automation, #Innovation, #TechTrends, #DigitalTransformation
- Select 3-5 most relevant hashtags
Determine risk level:
- low: Content < 200 chars, no links, matches approved topics
- medium: Content >= 200 chars OR contains links

Step 4: Create LinkedIn Approval Request

Generate approval request file in /Pending_Approval/:

---
type: approval_request
action: linkedin_post
plan_id: /Plans/PLAN_{slug}.md
source_action_item: /Needs_Action/{original_file}.md
created: {ISO_TIMESTAMP}
expires: {ISO_TIMESTAMP + 24h}
status: pending
priority: medium
risk_level: low|medium
mcp_server: linkedin-mcp
mcp_tool: create_post
---

## LinkedIn Post Draft

**Content**:
{generated_post_content}

**Hashtags**: #AI #Automation #Innovation

**Visibility**: PUBLIC

**Character Count**: {count}/280

## Risk Assessment

- Risk Level: {low|medium}
- Reason: {explanation}
- Auto-Approval Eligible: {yes|no}

## Parameters

- **text**: {post_content_with_hashtags}
- **visibility**: PUBLIC
- **hashtags**: ["AI", "Automation", "Innovation"]

## To Approve

Move this file to `/Approved/` folder to publish on LinkedIn.

## To Reject

Move this file to `/Rejected/` folder.

Example LinkedIn Post Generation:

Input action item:

Subject: New AI feature launch announcement
Content: We've just released our new AI-powered automation feature that reduces manual work by 50%...

Generated LinkedIn post draft:

Excited to announce our new AI-powered automation feature that reduces manual work by 50%!

This is a game-changer for productivity and efficiency.

#AI #Automation #Innovation #Productivity

2.7 Archive Processed Items

After creating the plan (and approval request if needed):

Move processed file from /Needs_Action/ to /Done/
Add completion metadata to the moved file:
- Processed timestamp
- Plan file reference
- Approval request reference (if applicable)
- Processing notes

3. File Structure Requirements

3.1 Vault Structure (Bronze + Silver Tier)

Bronze Tier Minimum:

vault/
├── Dashboard.md              # System status (updated by this skill)
├── Company_Handbook.md       # Rules and guidelines (read by this skill)
├── Needs_Action/             # Input: New items to process
│   ├── EMAIL_*.md           # From Gmail watcher
│   └── FILE_*.md            # From filesystem watcher
├── Plans/                    # Output: Generated plans
│   └── PLAN_*.md
└── Done/                     # Archive: Processed items
    └── [original_filename]_[timestamp].md

Silver Tier Extended (includes Bronze +):

vault/
├── Dashboard.md              # System status
├── Company_Handbook.md       # Rules, MCP config, approval thresholds
├── Needs_Action/             # Input: New items to process
├── Plans/                    # Output: Generated plans
├── Pending_Approval/         # SILVER: Approval requests waiting for human review
│   └── APPROVAL_*.md
├── Approved/                 # SILVER: Human-approved actions ready for execution
│   └── APPROVAL_*.md
├── Rejected/                 # SILVER: Human-rejected actions archive
│   └── APPROVAL_*.md
├── Done/                     # Archive: Processed items
└── Logs/                     # SILVER: Mandatory audit logs
    └── YYYY-MM-DD.json

3.2 Action Item File Format

Action items in /Needs_Action/ should follow this structure:

---
type: [email|file_drop|manual]
from: [sender/path]
subject: [subject/title]
received: [ISO_TIMESTAMP]
priority: [high|medium|low|auto]
status: pending
---

## Content
[The actual content of the action item]

## Metadata
[Additional metadata from watcher]

4. Processing Workflow

Step-by-Step Process

Detect New Items
- Scan /Needs_Action/ for .md files
- Identify files that haven't been processed (check for existing plans)
Read and Analyze
- Read action item file
- Read Company_Handbook.md for context
- Analyze content and determine priority
Create Plan
- Generate structured Plan.md in /Plans/
- Use clear, actionable language
- Include checkboxes for next steps
Update Dashboard
- Read current Dashboard.md
- Update counters and recent activity
- Preserve existing sections while updating dynamic content
Archive
- Move processed file to /Done/
- Add processing metadata
Create Approval Requests (Silver Tier - if external actions needed)
- Check if plan requires external actions
- Create approval request in /Pending_Approval/ if needed
- Update plan to reference approval request
- DO NOT execute external actions directly
Log Activity (Bronze: Optional, Silver: Mandatory)
- Bronze: Basic logging to console or simple log file
- Silver: Structured JSON audit log in /Logs/YYYY-MM-DD.json
- Record: timestamp, action_type, item_processed, plan_created, approval_requested (if applicable)

5. Error Handling

Common Errors and Responses

Missing Company_Handbook.md:
- Log warning
- Process with default priority rules
- Suggest creating handbook in plan notes
Malformed Action File:
- Log error with filename
- Skip processing (don't crash)
- Create error notification in Dashboard
Vault Permission Issues:
- Log error
- Stop processing
- Alert user via Dashboard status
Duplicate Items:
- Detect by content similarity or metadata
- Skip duplicate
- Add note to Dashboard about duplicate detection

6. Tier Capabilities

Bronze Tier

✅ Can Do:

Read from vault
Write Plan.md files
Update Dashboard.md
Move files within vault
Analyze and prioritize

❌ Cannot Do:

Send emails automatically
Execute external actions
Create approval workflows
Interact with MCP servers for actions

Silver Tier (EXTENDS Bronze)

✅ Can Do (Bronze capabilities +):

Create approval request files in /Pending_Approval/
Determine if actions require approval based on Company_Handbook.md thresholds
Log all activities to structured audit logs (/Logs/YYYY-MM-DD.json)
Reference MCP server capabilities in plans
Create plans that specify which MCP tools should be used
Handle approval request lifecycle (pending → approved/rejected)

❌ Cannot Do:

Execute external actions directly (MUST go through approval workflow)
Call MCP servers without human approval
Skip approval workflow for sensitive actions
Auto-execute payments, email sends, or social media posts without approval

Note: Use the execute-approved-actions skill to handle approved actions and invoke MCP servers.

7. Testing the Skill

To test this skill:

Create test action item:

echo "---\ntype: manual\nsubject: Test Item\nreceived: 2026-01-09T10:00:00Z\nstatus: pending\n---\n## Test Content\nThis is a test action item." > vault/Needs_Action/TEST_item.md

Invoke Claude Code with prompt:

Process any new action items in /Needs_Action folder. Create plans and update dashboard.

Verify:
- Plan.md created in /Plans/
- Dashboard.md updated
- Test file moved to /Done/

8. Example Usage

User Prompt:

Check the Needs_Action folder and process any new items. Create plans for each one.

Skill Execution (Bronze Tier):

Reads /Needs_Action/EMAIL_12345.md (from Gmail watcher)
Reads Company_Handbook.md for email response rules
Creates /Plans/PLAN_email_response_2026-01-09.md with:
- Analysis of email request
- Recommended response steps
- Priority classification
Updates Dashboard.md with new pending plan count
Moves EMAIL_12345.md to /Done/

Skill Execution (Silver Tier - with External Actions):

Reads /Needs_Action/EMAIL_12345.md (from Gmail watcher)
Reads Company_Handbook.md for email response rules and approval thresholds
Creates /Plans/PLAN_email_response_2026-01-09.md with:
- Analysis of email request
- Recommended actions (including "Send email reply")
- Priority classification
Creates approval request /Pending_Approval/APPROVAL_email_reply_12345.md:
- Action: send_email
- Target: sender@example.com
- Parameters: subject, body, attachments
- References plan file
Updates plan to reference approval request and mark status as pending_approval
Logs approval request to /Logs/YYYY-MM-DD.json
Updates Dashboard.md with pending approval count
Moves EMAIL_12345.md to /Done/

Expected Output (Bronze):

One or more Plan.md files created
Dashboard.md reflects new activity
Action items archived to /Done/

Expected Output (Silver):

Plan.md files created with external action recommendations
Approval request files in /Pending_Approval/ (for actions requiring approval)
Audit log entries for approval requests
Dashboard showing pending approvals
Action items archived to /Done/
Human reviews and approves, moves approval to /Approved/, then @execute-approved-actions skill executes

9. Best Practices

Do:

Always read Company_Handbook.md before processing
Create clear, actionable plans with checkboxes
Preserve original content in plans (reference original file)
Update Dashboard incrementally (don't overwrite user customizations)
Handle errors gracefully (skip bad items, don't crash)

Don't:

Modify original action files (only move them)
Hardcode processing rules (always reference Company_Handbook.md)
Create plans for items already processed (check duplicates)
Assume file formats (handle missing metadata gracefully)
Execute external actions (Bronze tier is read-only)

By following this skill, you act as a reliable action item processor:

Converting watcher discoveries into actionable plans,
Maintaining system visibility through Dashboard updates,
Organizing workflow through proper file management,
And establishing the foundation for autonomous operation in higher tiers.

process-action-items

SKILL.md