| name | find-duplication |
| description | Find code duplication in the codebase. Supports two modes - scoped to current branch changes or a full codebase sweep. Use when the user asks to find duplicated code, copy-paste, repeated patterns, or wants to deduplicate before a PR. |
| disable-model-invocation | true |
Find Code Duplication
Detect duplicated or near-duplicate code and suggest consolidation candidates.
Rules
- Report findings, do not refactor. Refactoring is a separate task.
- Focus on production code (
ols/). Skip test duplication unless explicitly asked.
- Group findings by severity: exact duplicates first, then near-duplicates.
- For each finding, state whether extraction is worth it or acceptable duplication.
Step 1: Determine Scope
Ask the user:
- Branch mode: only files changed in the current branch vs main.
- Full mode: scan the entire
ols/ directory.
For branch mode:
git diff --name-only origin/main -- 'ols/' | grep '\.py$'
For full mode, the target is ols/.
Step 2: Run Pylint Duplicate Detection
uv run pylint --disable=all --enable=duplicate-code --min-similarity-lines=6 <target files or directory>
Review output. Filter out false positives:
- Import blocks (common imports are not duplication)
- Pydantic model boilerplate (Field declarations)
- Single-line patterns (logging, raises)
Step 3: Semantic Duplication Search
Pylint only catches textual similarity. Also look for:
- Similar function signatures — functions with near-identical parameter lists doing similar work.
- Repeated error handling — same try/except/log/return pattern across multiple files.
- Copy-pasted blocks — search for distinctive string literals or variable names that appear in multiple files.
rg "<distinctive pattern>" ols/ --type py -l
Step 4: Classify Findings
For each duplicate found, classify:
| Category | Action |
|---|
| Extract — identical logic in 3+ places | Recommend a shared helper |
| Parameterize — same structure, different values | Recommend a common function with parameters |
| Acceptable — similar but serving different domains | Note it, no action needed |
| Test-only — repeated test setup/fixtures | Recommend shared fixture (only if user asked) |
Step 5: Report
For each finding:
- Files and line ranges involved
- What is duplicated (brief description)
- Classification (extract / parameterize / acceptable)
- Suggested location for shared code (if applicable)
Summary: total findings, how many actionable, estimated lines saved.
