Creating Your Own Thinking Tools

How to design and build custom thinking tools for your specific domains and thinking patterns.

Anatomy of a Thinking Tool

A thinking tool is a slash command (.md file in ~/.claude/commands/) with specific design principles that make it effective for reasoning, not just task execution.


  ┌─────────────────────────────┐
  │  Trigger                    │
  │  User invokes /tool-name    │
  └──────────────┬──────────────┘
                 │
                 ▼
  ┌─────────────────────────────┐
  │  Context Load               │
  │  Read relevant vault files  │
  └──────────────┬──────────────┘
                 │
                 ▼
  ┌─────────────────────────────┐
  │  Prompt Frame               │
  │  Set the reasoning mode     │
  └──────────────┬──────────────┘
                 │
                 ▼
  ┌─────────────────────────────┐
  │  Interaction                │
  │  Guided dialogue or analysis│
  └──────────────┬──────────────┘
                 │
                 ▼
  ┌─────────────────────────────┐
  │  Output                     │
  │  Insights, challenges, plans│
  └──────────────┬──────────────┘
                 │
                 ▼
  ┌─────────────────────────────┐
  │  Capture                    │
  │  Save to vault (optional)   │
  └─────────────────────────────┘

Design Principles

1. Context-Aware

Good thinking tools don’t operate in a vacuum. They read from the vault to understand your current state.


# Bad: Generic
"Give me some creative ideas about this topic."
 
# Good: Context-aware
"Read my current goals from [goals file]. Read my recent
conversations from [session logs]. Given this context,
surface patterns I might be missing."

2. Vault-Connected

Thinking tools should read from and optionally write to the vault. This creates a feedback loop where past thinking informs future thinking.

Action	When
Read vault files	Always — context improves output quality
Write to vault	When the output should persist (reflections, decisions, patterns)
Skip writing	For quick challenges or brainstorming where capture isn’t needed

3. Reflective, Not Prescriptive

Thinking tools surface information and perspectives. They don’t make decisions for you.


# Bad: Prescriptive
"You should definitely use React for this project because..."
 
# Good: Reflective
"Here are 3 trade-offs to consider:
1. React gives you X but costs Y
2. Vue gives you A but costs B
3. Your past projects suggest you value Z
What matters most here?"

4. Mode-Setting

The prompt should explicitly set a reasoning mode for the AI:


# Devil's advocate mode
"Take the opposite position from whatever I say.
Challenge every assumption. Be rigorous but constructive."
 
# Pattern-matching mode
"Look across multiple domains and time periods.
Find recurring themes, not just surface similarities."
 
# Divergent thinking mode
"Deliberately move away from obvious solutions.
Use analogies, metaphors, and cross-domain connections."

Template: Basic Thinking Tool


# Tool Name
 
Brief description of the thinking mode this tool activates.
 
## Context Loading
Read the following files for context:
- [relevant vault file 1]
- [relevant vault file 2]
 
## Prompt Frame
[Set the AI's reasoning mode here]
 
## Interaction Pattern
1. Ask the user to describe their situation/question
2. [Apply the thinking mode]
3. Present findings as observations, not directives
4. Ask a follow-up question to deepen the thinking
 
## Output
[What format the response should take]
 
## Capture (Optional)
$BASH
# Save insights to vault if valuable
echo "## [Date] - Tool Name Session" >> ~/vault/reflections.md
$ENDBASH

Template: Advanced Thinking Tool (Multi-Step)


# Tool Name
 
Multi-step thinking process for [purpose].
 
## Step 1: Gather Context
$BASH
# Read relevant files
cat ~/vault/goals.md
cat ~/vault/recent-sessions.md
$ENDBASH
 
## Step 2: Analyze
Based on the context above, [specific analysis instruction].
 
Present your analysis as:
- **Patterns observed:** [list]
- **Contradictions found:** [list]
- **Questions raised:** [list]
 
## Step 3: Challenge
Now take the opposite position. What would someone who
disagrees with these patterns say?
 
## Step 4: Synthesize
Combine the analysis and challenges into 2-3 key insights
that are actionable.
 
## Step 5: Capture
$BASH
# Optional: save to vault
$ENDBASH

Example: Custom Domain Tool

Here’s an example of a thinking tool for investment decisions:


# /invest-think
 
Investment decision analysis tool.
 
## Context
Read current portfolio from portfolio.md.
Read investment thesis from thesis.md.
Read recent market notes from market-log.md.
 
## Frame
You are a rigorous investment analyst. Your job is NOT to
validate the user's existing bias but to stress-test the
thesis. Apply these lenses:
1. What's the bear case?
2. What information would change this thesis?
3. What's the opportunity cost vs current holdings?
4. What does the position size say about conviction?
 
## Interaction
1. User describes the investment opportunity
2. Apply all 4 lenses
3. Present a structured analysis
4. End with: "What would make you NOT do this?"

Anti-Patterns

Don’t	Do Instead
Make the tool give direct advice	Make it ask better questions
Skip context loading	Always read relevant vault files
Output walls of text	Use structured formats (tables, bullet lists)
Make it generic (“brainstorm anything”)	Give it a specific reasoning mode
Forget to capture insights	Add optional vault writing for valuable sessions

Testing Your Tool

Before adding a new thinking tool to your workflow:

Use it 3 times on real problems (not test cases)
Check the output — is it genuinely useful or just verbose?
Compare to raw prompting — does the tool add value over just asking Claude directly?
Get feedback — if others use it, do they find it valuable?

If a tool doesn’t pass these checks, it’s not ready. Iterate on the prompt frame and context loading until it consistently produces useful thinking.