glebis/thinking-patterns

Thinking Patterns Skill

Longitudinal cognitive pattern analysis across months of recorded conversations. Extracts evidence-based dimensions from Fathom transcripts, synthesizes cross-session patterns, and detects blind spots you can't see yourself.

Usage

/thinking-patterns                              # full analysis, default period (3 months)
/thinking-patterns --dry-run                    # corpus stats + batch plan only
/thinking-patterns --period 2026-01 2026-02     # custom date range

What it does

1. Discovers all Fathom transcripts in date range, classifies by type (coaching, meeting, lab, etc.)
2. Extracts Gleb's speech lines from each transcript
3. Analyzes 12 evidence-based cognitive dimensions per transcript (parallel sonnet agents)
4. Synthesizes cross-session patterns across 10 analysis sections (parallel sonnet agents)
5. Detects contradictions, competing commitments, and blind spots
6. Produces single analysis document with dated evidence quotes

12 Extraction Dimensions

| Dimension | Framework | |-----------|-----------| | Cognitive distortions | Burns' 10 categories | | Problem framing | Discourse analysis | | Conceptual metaphors | Lakoff & Johnson / MIPVU | | Hedging & certainty | Epistemic marker taxonomy | | Code-switching | Bilingual cognition research | | Decision moments | GDMS + behavioral markers | | Emotional indicators | Russell's Circumplex + ACT | | Avoidance & deflection | ACT experiential avoidance | | Agency language | Narrative Identity Theory | | Competing commitments | Immunity to Change (Kegan) | | Role/register markers | Sociolinguistic analysis | | Energy signals | Engagement markers |

10 Output Sections

1. Recurring Narratives
2. Problem Framing Patterns
3. Metaphors & Unconscious Language
4. Decision Heuristics
5. Topics Avoided
6. Contradictions & Competing Commitments
7. Energy Patterns
8. Role Shifts
9. Execution Gap (stated priorities vs actual focus)
10. Cognitive Distortions & Biases

Plus: "The 5 Things You Don't See" summary.

Output

• Analysis: ai-research/YYYYMMDD-thinking-patterns-analysis.md
• Daily note: Link appended under ## Research

Architecture

Stage 0: Corpus Discovery (~30s, orchestrator)
  ├── Find transcripts, classify, extract Gleb's speech
  └── Load reference docs, plan batches

Stage 1: Per-Transcript Extraction (~3 min, 10-13 parallel sonnet agents)
  ├── Each agent processes 1-4 transcripts
  └── Returns structured JSON per transcript (12 dimensions)

Stage 2: Aggregation (~30s, orchestrator)
  ├── Parse JSONs, de-duplicate, cluster
  └── Package into 5 synthesis bundles

Stage 3: Cross-Session Synthesis (~2 min, 5 parallel sonnet agents)
  ├── Narratives + Metaphors + Language
  ├── Decisions + Problem Framing + Biases
  ├── Avoidance + Energy + Execution Gap
  ├── Role Shifts + Developmental Markers
  └── Blind Spot Detection

Stage 4: Output (~1 min, orchestrator)
  └── Compile, save, link

Cost

~$3.50 per full run, ~6 minutes runtime.

Requirements

• Fathom transcripts in vault root (YYYYMMDD-*.md format)
• Reference docs: Profile Brief, My Focus, Strategic Decisions Framework
• No external dependencies -- pure Claude Code tools