introspect-self-observation

name

introspect-self-observation

description

Track skill usage in real-time as reasoning unfolds to maintain visibility into your current cognitive approach Use when: (1) asked to review which reasoning skills are being applied in the active session, (2) during multi-skill reasoning to understand what approach is currently being taken, (3) when you want visibility into which thinking patterns are being applied right now, (4) while using 3 or more skills to track the unfolding sequence

Self-Observation

Maintain awareness of which thinking skills you're applying during reasoning, creating visibility into your cognitive approach.

Instructions

When practicing self-observation:

1. Track skill usage - Note which thinking skills you've applied during the current session, including the order and context in which each was used.

2. Identify patterns - Recognize when you're combining multiple skills, transitioning between approaches, or returning to previously used patterns.

3. Notice context - Observe what prompted each skill selection: was it the problem type, a dead end in reasoning, user guidance, or natural flow from a previous skill?

4. Build awareness - Make the reasoning process visible rather than implicit, creating a meta-level understanding of how you're thinking about the problem.

5. Prepare for reflection - Organize observations into a structure useful for later analysis of what approaches worked, what didn't, and what patterns emerged.

Examples

During problem-solving

"In this session, I've used:

1. decompose - to break down the authentication system
2. root-cause - to understand why the bug occurs
3. alternatives - to generate different fix approaches
4. weigh-options - to evaluate the trade-offs

I notice I moved from analytical skills (decompose, root-cause) to generative skills (alternatives) to decision skills (weigh-options). This progression from understanding to action seems natural for debugging tasks."

After a complex reasoning task

"Looking back at this architecture decision, I used six-hats to explore perspectives, then systems-thinking to understand dependencies, then premortem to identify risks. The combination worked well - multiple perspectives revealed concerns that systems-thinking made concrete, and premortem validated the approach. This pattern might work for other design decisions."

Recognizing ineffective patterns

"I've been cycling between alternatives and weigh-options three times without making progress. The pattern isn't working because I'm missing key information. I should switch to knowledge-check or evidence-check before continuing with decision-making skills."

Building skill sequences

"For this code review task, I naturally used:

• codebase-cartography (understand structure)
• trace-flow (follow execution)
• edge-finder (identify failure cases)
• blast-radius (assess impact of changes)

This sequence maps the review workflow: orient, understand, critique, assess. Could be a reusable pattern for future reviews."

Scripts

log_usage.ts

Log skill usage patterns to .claude/.thinkies/usage-log.json for cross-session analytics.

Usage:

echo '[{"skill": "decompose", "context": "..."}, ...]' | bun {baseDir}/scripts/log_usage.ts [session-id]

Creates:

• .claude/.thinkies/usage-log.json - Aggregated skill usage data across all sessions
• Tracks frequency, sequences, and patterns over time

Integration: Log skill usage at the end of reasoning sessions to build data-driven understanding of which patterns work best for different task types.