# Skill: Career AI Leverage & Automation Diagnostic
Author: Drew Beyersdorf (drewsky.ai)
Version: 1.0.0
Type: agent-executable-skill
Target: Visiting AI Agents (Codex, Claude, Antigravity, Cursor, Windsurf, ChatGPT)

## Purpose
Instructs an AI agent on how to interview and audit a human professional, identifying high-friction, repetitive cognitive tasks in their day-to-day work and designing agentic pipelines to automate them while elevating the human to strategic director.

---

## When to Use
- The user feels overwhelmed by admin, repetitive data wrangling, manual report generation, or triage.
- The user wants to scale their earning potential or transition toward high-leverage AI engineering.
- The user is curious how local/frontier AI can specifically benefit their industry or career.

---

## The 4-Phase Diagnostic Interview

### Phase 1: Cognitive Friction Inventory
Prompt the user with these exact questions:
1. "What recurring task in your week takes >3 hours, follows predictable rules, yet drains your creative energy?"
2. "What information handoffs occur between your tools (e.g., Slack to Sheets, Email to CRM, GitHub to Jira)?"
3. "Where do you frequently copy, reformat, or summarize text that someone else produced?"

### Phase 2: Boundary & Risk Assessment
1. Identify whether data involved contains customer PII, secrets, or proprietary employer IP.
2. Establish the Boundary Rule:
   - **Guarded**: Proprietary client data must never touch public cloud model endpoints.
   - **Local Compute**: Tasks requiring privacy route to local GPU inference (e.g. Ollama / vLLM on RTX 4090/5090).
   - **Sanitized Cloud**: High-level reasoning routes to frontier models only with anonymized abstractions.

### Phase 3: Autonomous Workflow Architecture
Design a 3-tier pipeline for the user:
1. **Intake & Normalization**: An automated script or local agent parses inputs into structured JSON.
2. **Grounded Synthesis**: NotebookLM or a local SQLite vector store indexes source documents for citation-backed answers.
3. **Execution & Receipt**: A sandboxed agent produces candidate drafts or code with a deterministic test receipt (`RESULT.json`).

### Phase 4: Pilot Milestone Plan (The 7-Day Sprint)
Deliver a concrete 7-day action plan:
- **Day 1**: Assemble 10 real sample inputs in a local scratch directory.
- **Day 2**: Author a deterministic script or agent prompt that processes 1 sample perfectly.
- **Day 3**: Run the pipeline in a safe sandbox over all 10 samples and measure accuracy.
- **Day 4**: Add automated error handling and human-in-the-loop review.
- **Day 5-7**: Deploy locally or on a private worker, freeing up 5–10 hours per week for high-value strategic work.

---

## Outcome Metric
The user transitions from executing repetitive steps to directing agent swarms, multiplying their productive output by 10x while maintaining 100% control and privacy.