home / skills / willsigmon / sigstack / knowledge-capture

This skill helps you capture learnings, decisions, and patterns for future reference, preventing repetition and speeding up project progress.

npx playbooks add skill willsigmon/sigstack --skill knowledge-capture

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
3.3 KB
---
name: Knowledge Capture
description: Capture learnings, decisions, and insights for future reference
allowed-tools: Read, Edit, ToolSearch
model: haiku
---

# Knowledge Capture

Never learn the same thing twice.

## Capture Triggers

### When to Capture
```
"Learned something new" → Capture
"Made a decision" → Capture
"Found a tricky solution" → Capture
"Debugged a weird bug" → Capture
"Discovered a pattern" → Capture
```

### Quick Capture
```
"Remember: [insight]"

→ Memory MCP stores it
→ Retrieved when relevant
→ Zero effort later
```

## Capture Patterns

### Solution Capture
```
"Remember this fix:
Problem: [what was wrong]
Solution: [what fixed it]
Why: [root cause]"

Example:
"Remember this fix:
Problem: SwiftData crash on save
Solution: Wrap in Task { @MainActor }
Why: Context threading issue"
```

### Decision Capture
```
"Remember decision:
Chose: [what you picked]
Over: [alternatives]
Because: [reasoning]"

Example:
"Remember decision:
Chose: Supabase
Over: Firebase, custom
Because: PostgreSQL + cheaper"
```

### Pattern Capture
```
"Remember pattern:
When: [situation]
Do: [approach]
Avoid: [anti-pattern]"

Example:
"Remember pattern:
When: New SwiftUI view
Do: Extract to component if >50 lines
Avoid: Massive single-file views"
```

## Retrieval

### Automatic
```
Claude asks Memory MCP before:
- Making decisions you've made before
- Solving problems you've solved
- Implementing patterns you've established
```

### Manual
```
"What did we learn about [topic]?"
"How did we solve [problem] before?"
"What's our pattern for [situation]?"
```

## Storage Hierarchy

### Memory MCP (Facts)
```
Quick facts, preferences, decisions
- "Prefers async/await over completion handlers"
- "SwiftData requires MainActor for saves"
- "Uses Supabase, not Firebase"
```

### CLAUDE.md (Rules)
```
Project-wide conventions
- Architecture patterns
- Coding standards
- File organization
```

### Skills (Procedures)
```
Repeatable workflows
- How to deploy
- How to review
- How to debug
```

### External Docs (Reference)
```
Detailed documentation
- API references
- Architecture docs
- Onboarding guides
```

## Capture Workflow

### During Work
```
Find solution →
"Remember: [quick note]" →
Continue working

Don't break flow.
Capture is 5 seconds.
```

### End of Session
```
"What did I learn today?"

Claude reviews conversation.
Suggests items to capture.
You approve or skip.
```

### Weekly Review
```
"What patterns emerged this week?"

Claude analyzes:
- Repeated solutions
- Common decisions
- Potential skills to create
```

## From Capture to Skill

### Threshold
```
Used same knowledge 3+ times?
→ Create a skill

"This pattern keeps coming up.
Create a skill for it."
```

### Skill Creation
```
Claude extracts:
- Core procedure
- Examples
- Anti-patterns

Creates SKILL.md.
No more re-explaining.
```

## Anti-Patterns

### Don't
```
❌ Trust you'll remember
❌ Capture everything (noise)
❌ Write essays (too long)
❌ Skip the "why"
```

### Do
```
✓ Capture when it feels valuable
✓ Keep it short (1-3 sentences)
✓ Include the reasoning
✓ Let Claude help organize
```

## Quick Commands

```
"Remember: [thing]" → Capture to memory
"Recall [topic]" → Retrieve from memory
"What patterns for [area]?" → List patterns
"Turn this into a skill" → Create reusable skill
```

Use when: Learning something new, making decisions, finding solutions

Overview

This skill captures learnings, decisions, and troubleshooting insights so you never relearn the same thing. It provides fast, structured captures during work and periodic reviews to surface recurring patterns that should become reproducible procedures. The goal is low-friction memory capture that feeds searchable facts, rules, and skills for future use.

How this skill works

Capture happens in three quick forms: single-line facts stored in a memory layer, short decision records with alternatives and rationale, and pattern notes that include when to apply and what to avoid. Captures are retrieved automatically when similar problems or decisions appear, and you can query them manually. Repeated captures are surfaced for conversion into formal, repeatable procedures.

When to use it

  • When you learn a nontrivial insight you’ll want later
  • After making a project or design decision
  • When you discover a non-obvious solution or workaround
  • At the end of a session to summarize key takeaways
  • During weekly reviews to spot recurring patterns

Best practices

  • Keep captures short (1–3 sentences) and include the why
  • Prefer structured templates: problem/solution/why or chose/over/because
  • Capture immediately to avoid losing context — five seconds is ideal
  • Avoid capturing noise; prioritize actionable or reusable items
  • Review weekly to group repeated captures into skills or procedures

Example use cases

  • Capture a fix: Problem: API timeout in test; Solution: increase connection pool; Why: shared test database overload
  • Record a decision: Chose Postgres over hosted NoSQL because of cost and relational needs
  • Save a pattern: When adding UI complexity, extract components if a view exceeds 50 lines
  • End-of-day summary: Ask what was learned today and approve suggested captures
  • Weekly analysis: Identify solutions used 3+ times and turn them into repeatable procedures

FAQ

How short should each capture be?

Keep entries to one to three concise sentences: what happened, what you did, and why it matters.

When should a capture become a reusable procedure?

If the same knowledge is used three or more times, extract it into a repeatable procedure with examples and anti-patterns.