find-similar

Find Similar Patterns

A conceptual framework for systematically finding similar or analogous code patterns in a codebase.

This skill is factual, not prescriptive. It helps find and report what exists. It does not recommend whether to use, ignore, or modify findings.

Tools available: Grep, Glob, Read, Bash (git commands)

---

Similarity Types

"Similar" can mean different things. Identify which type matters before searching.

Type	What It Means	Example
Lexical	Same names, keywords, identifiers	"Where else do we call formatDate?"
Structural	Same code shape, different names	"Where else do we have retry logic?"
Analogous	Same role in a different domain	"What's the equivalent handler in another domain?"
Conceptual	Same purpose, potentially different approach	"How do we handle validation elsewhere?"

---

Search Strategy

Level 1: Direct Search

Search for the thing itself or obvious variations.

• Exact terms, function names, type names
• Known synonyms or alternate spellings
• Import/export statements

Stop if: Found clear matches.

Level 2: Sibling/Peer Discovery

Find files that serve the same role.

• Files in the same directory
• Files with the same naming pattern (e.g., *.handler.ts, use*.ts)
• Files in parallel directories (e.g., domains/users/ → domains/projects/)

Stop if: Found peers that reveal the pattern.

Level 3: Reference Tracing

Follow the dependency graph.

• Where is X defined?
• What imports/uses X?
• What does X import/use?

Stop if: Found the relevant connected files.

Level 4: Conceptual Expansion

Broaden the search with related concepts.

• Synonyms and related terms
• Different implementations of the same idea
• Cross-domain analogues

Stop if: Found conceptually similar code, or exhausted reasonable search terms.

---

Confidence Levels

Confidence	Criteria
HIGH	Exact or near-exact match; clearly the same pattern
MEDIUM	Similar structure or purpose; some differences
LOW	Conceptually related; different approach or partial match

Factors that affect confidence:

• Same directory/domain → higher
• Same naming conventions → higher
• Same imports/dependencies → higher
• Different structure or approach → lower

---

What to Capture (adapt to your context)

Useful information to track for each finding:

• Location — file path and line range
• Similarity type — which of the four types applies
• Confidence — how close is the match
• Why similar — brief explanation of the relationship

For negative results, note what was searched so coverage can be verified.

---

Tips

• Start narrow, expand as needed — Don't search the entire codebase if a directory search suffices
• Use file organization as signal — Sibling files often reveal local conventions
• Git history can help — Files that change together are often related
• Report what you searched — Helps verify coverage and enables follow-up