blind-spot-finder— agent skill

Use multiple AI models to find what I'm NOT seeing. This is about the unknown unknowns.

Step 1: Gather Context

Pull data in parallel using LifeOS MCP tools:

1. get_working_memory — ALL sections
2. get_coaching_action_items with status "pending" — incomplete homework
3. get_habits — habits and their completion rates
4. get_screentime_summary for the last 3 days
5. get_health_sleep for the last 14 days
6. get_finance_net_worth — financial position

Step 2: Build the Blind Spot Brief

Compile a comprehensive brief of my current situation including:

• Working memory patterns and triggers
• What I say I want vs what my data shows I actually do
• Action items I keep not completing
• Screen time vs stated priorities
• Any say-do gaps visible in the data

Step 3: Multi-Model Council

Use the zen consensus tool with this prompt structure:

Round 1 — Blind Spot Detection: "Here is a detailed profile of a person: [insert brief].

Identify the top 3-5 blind spots this person likely has. Focus on:

• Self-deceptions they're maintaining
• Local maxima they're stuck in
• Assumptions they haven't questioned
• Patterns they can't see because they're inside them
• Things they SAY they want but systematically avoid

Be brutally honest. This person explicitly wants to be challenged, not comforted."

Round 2 — Pattern Challenge: Use the zen challenge tool: "This person has identified these patterns about themselves: [insert patterns from working memory].

Challenge each pattern: Is their self-diagnosis accurate, or is THAT diagnosis itself a blind spot? What if the real issue is something they haven't named? Propose alternative framings that might unlock new action."

Step 4: Synthesize

Present the findings organized as:

Confirmed Blind Spots (models agree)

Things multiple models flagged — these are high-confidence insights.

Contested Insights (models disagree)

Where models disagreed — present both sides. The disagreement itself is informative.

The Uncomfortable Questions

Synthesize 3-5 questions that emerged from the analysis. These should be questions that make me squirm — that's how I know they're hitting something real.

Reframes

For each major blind spot, offer an alternative way to see the situation. Not advice — just a different lens.

Step 5: Choose One to Act On

Ask: "Which of these landed? Pick ONE blind spot to work on this week."

Then create a coaching action item with create_coaching_action_item — a specific experiment to test whether this blind spot is real.

Update working memory with update_working_memory if any genuinely new insight emerged.