Dart Dedupe (dart-dedupe)

High-performance structural code duplication and clone detection engine and refactoring protocol for Dart and Flutter repositories using pkg:dedupe.

1. When to Use This Skill

Use this skill when auditing codebase redundancy, identifying copy-paste blocks, preventing duplication regressions in pull requests, or evaluating shared utility and abstraction candidates across Dart packages.

Unlike basic lexical diffs, dedupe tokenizes and analyzes Dart source code to detect:

• Identical Clones: Exact token-for-token copies across files.
• Structural Clones: Clones matching after normalizing string literals and

numeric constants (--ignore-literals).

• Parameterized Clones: Clones matching after normalizing variable and type

identifiers (--ignore-identifiers).

• Gapped / Near-Miss Clones: Type-3 clones matching with internal statement

insertions, deletions, or edits via MinHash & LSH shingling (--bucket=gapped).

Trigger Indicators

• Copy-pasted utilities: Identical or nearly identical helper functions

duplicated across multiple files or classes.

• Redundant serialization / parsing: Repeated manual JSON decoders, record

mappers, or data transformation pipelines.

• PR / CL Duplication Regressions: Catching newly introduced duplicate code

before merging pull requests.

When NOT to Use

• Single-File Private Variable Lints: Use standard dart analyze for simple

unused variables or parameters.

• Code Formatting: Use dart format.
• Non-Dart Projects: Tool operates strictly on Dart syntax.

2. Automated Execution & Scope Resolution

Run the official package CLI directly:

dart run dedupe@^0.1.0 [options] [target_path]

Execution Modes

Mode 1: Full Repository / Directory Scan

# Markdown summary with clickable file links
dart run dedupe@^0.1.0

# Machine-readable JSON output for agent pipelines
dart run dedupe@^0.1.0 --format=json

# Write JSON report to file alongside human stdout
dart run dedupe@^0.1.0 --json-output=report.json

Mode 2: PR / Git Diff Delta Scan (--git-diff)

Focus strictly on code modified in a branch or PR:

# In Git checkouts:
dart run dedupe@^0.1.0 --git-diff=origin/main

# Filter report strictly to clusters intersecting modified lines:
dart run dedupe@^0.1.0 --git-diff=origin/main --only-changed

# Fail CI if diff duplication exceeds 5%:
dart run dedupe@^0.1.0 --git-diff=origin/main --fail-threshold=5

Common CLI Options Reference

<!-- mdformat off(prevent table wrapping) -->

<!-- mdformat on -->

3. The "Actionable vs. Necessary" Architectural Gate

Do not treat every duplicate finding as a mandatory refactoring target. Evaluate each candidate cluster before modifying code:

✅ Actionable Duplication (Refactor & Extract)

• Copy-pasted helpers or decoders: Identical algorithms, database record

parsers, or conversion utilities scattered across multiple classes or files.

• Shared contract declarations: Common typedef contracts, data models, or

record shapes duplicated across platform stubs (extract to a shared library).

• Repetitive CLI orchestration: Copy-pasted external process invocations or

JSON decoding blocks where schema updates would risk drift.

• Code generator scaffolding: Repetitive string builders or verbose AST

instantiations that can be cleanly condensed into parameterized emitters.

🛑 Necessary Duplication (Reject Refactoring & Preserve)

• Type-unsafe polymorphic AST targets: When similar-looking classes (such as

AST statement variants) do not share a common type interface. Using dynamic sacrifices compile-time type safety for negligible line reduction.

• Performance-critical specialized solver loops: Symmetric horizontal vs.

vertical grid traversals where unifying orthogonal strides into a single abstraction would require allocating closures or virtual calls in tight loops.

• Speculative wrapping of standalone entry points: Abstracting trivial

4-to-6 line try/catch fallback formatting across unrelated standalone CLI entry points (bin/<script>.dart).

• Code generator template strings: Repetitive structural patterns in

multiline source code generation templates (e.g. build_runner code emitters).

• DAMP Test Boilerplate: Repetitive setup structures in table-driven unit

tests.

Invariant: Exhaustive Cluster Adoption

When extracting a shared helper, mixin, or base class for an actionable duplicate cluster:

• Mandatory Full-Cluster Scope: Within a target package or subsystem, you

MUST apply the shared abstraction across ALL participating files in that cluster in the same refactoring pass. Do not refactor only 1 file while leaving the remaining N - 1 files with duplicate logic.

• Subsystem Consolidation: If multiple sibling classes share multiple

duplicate clusters (e.g. description formatting, error handling, or stream lifecycle), design a unified shared base class or mixin consolidating all shared clusters at once.

• Encapsulation & Directory Placement: Place extracted internal shared

abstractions under lib/src/ (e.g. lib/src/shared/ or lib/src/common/). Never place internal helpers directly in the public lib/ root.

4. The 2-Stage Triage & Confirmation Protocol

To prevent unwanted diff bloat and preserve codebase stability, adhere to this strict 2-stage workflow:

Stage 1: Read-Only Audit & Reporting (Mandatory Stop)

Run dart run dedupe@^0.1.0 --format=markdown (or --format=json).

Output a ranked Duplication Triage Report containing:

1. Target Summary: Files analyzed, total lines, duplication percentage, and

estimated lines saved.

1. Top Duplicate Clusters: Clickable file links, line ranges, token counts,

clone classification category (logic, data, boilerplate), and bucket (identical, structural, parameterized, gapped).

1. Actionability Annotations: Highlight recommended extraction strategy or

mark as "Necessary Duplication (Preserve)".

1. Prioritization: Rank by highest token volume and widest file footprint

first.

Stage 2: Interactive User Confirmation Gate

Pause execution and prompt the user (via interactive choice or chat) to select the desired remediation scope:

1. (Recommended) Refactor Top Actionable Cluster Across All Files:

Consolidate the highest-impact duplicate cluster and refactor 100% of the participating files in that cluster.

1. Subsystem Base Refactor: Consolidate multiple related duplicate clusters

across sibling classes into a unified shared base/mixin.

1. Report-Only / Exit: Acknowledge findings without code mutations.

Explicit Bypass & Non-Interactive Fallback:
- Direct Directives: Skip Stage 1 pause if given explicit remediation
instructions (e.g., "Deduplicate clusters in pkgs/foo using dart-dedupe").
- Non-Interactive Execution: In unattended or automated evaluation
workflows (e.g. evalin or subagents), proceed with Option 1 (Refactor
Top Actionable Cluster Across All Files) automatically after verifying
baseline tests pass.

5. Pre & Post Refactor Verification Protocol

Wrap all deduplication refactoring in a strict verification sandwich:

1. Pre-Flight Baseline:

- Check pubspec.yaml: if sdk: flutter is declared, run flutter test; otherwise run dart test. - Confirm test suite is 100% green before touching code.

1. Surgical Modification: Extract shared functions or helper classes under

lib/src/ cleanly.

1. Verify Post-Refactor Health:

- Run flutter analyze or dart analyze --fatal-infos. - Run flutter test or dart test. - Monorepo Downstream Gate: In multi-package workspaces, run tests across all dependent packages.

1. Zero-Clone Confirmation: Re-run dart run dedupe@^0.1.0 to confirm the

target cluster was eliminated.

1. Local Staging: Stage verified diffs locally (git add .).

6. Pull Request & Commit Provenance Protocol

When staging deduplicated code and preparing a commit message or Pull Request:

1. User Confirmation Gate

• Interactive Sessions: Before writing the PR description or commit body,

explicitly prompt the user in chat or via the harness confirmation tool (e.g. ask_question) whether to include a Tool Provenance & Reproduction block.

• User Prompt Inclusion: When the user explicitly requests inclusion (or

confirms via prompt), append the standardized markdown block below. In unattended or automated workflows, output the summary to chat or step summaries rather than modifying commit bodies without user confirmation.

2. Standardized Provenance Block Format

When confirmed by the user, include the following markdown block in the PR description or commit body:

````markdown

🤖 Tool Provenance & Reproduction

Structural duplication analysis performed with dedupe (v{version}).

To reproduce or re-run this duplication scan locally:

{exact_command_line}

````

3. Version Resolution

Determine the package version dynamically:

• Check pubspec.lock in the workspace or run

dart run dedupe@^0.1.0 --version.

• If invoked with a specific version constraint (e.g. dedupe@^0.1.0), use that

exact version.