Roadmap research, September 2026: 20 feature issues ranked by demand and competitor gap

[kucherenko]

Summary

Tracking issue for the feature research pass done in September 2026 against jscpd v5.1.2. Inputs: all 375 jscpd issues and 10 open PRs, ~100 external GitHub issues that mention jscpd, ~30 competing and adjacent clone detectors (PMD CPD, Simian, SonarQube, JetBrains/Qodana, Qlty, mizchi/similarity, boyter/dcd, CodeClone, fossil-mcp, NiCad, SourcererCC and others), Hacker News / dev.to / Stack Overflow threads, npm download data, and the GitClear / Sonar / HackerNoon studies on duplication in AI-generated code.

Ranking uses three signals: how many people ask, whether competitors already have it, and how clear the gap is after v5 (baseline gating, MCP, SARIF/codeclimate/openmetrics, summary, blame and isolation groups are already shipped and were not counted as gaps).

Tier 1 — strong demand, clear gap, competitors have it

• Type-2 clone detection: normalize identifiers, literals and annotations (--ignore-identifiers, --ignore-literals, --ignore-annotations) #998 Type-2 detection: normalize identifiers, literals, annotations (refs Implement multiple algorithms to choose from besides Karp Rabin #462, Pre-process lines being tokenized. #489, how to use modes with API calls ? #577)
• Type-3 near-miss clones: gap-tolerant window merging and function-level similarity for JS/TS #999 Type-3 near-miss clones: gap-tolerant merging, function-level similarity via oxc (refs Implement multiple algorithms to choose from besides Karp Rabin #462, How do I find locations and snippets in a folder that are similar to the specified code snippet? #727, Pre-process lines being tokenized. #489)
• Clone classes: group N-way duplicates into one entry, deterministic ordering, --min-instances #1000 Clone classes: N-way grouping, deterministic output, --min-instances (refs Same code on multiple files #221 — most-upvoted open issue, Ability to support more than two clones per file, or across files?  #474)
• --watch and --cache: incremental re-scan on edit, reusing the MCP in-memory index #1001 --watch and --cache: incremental re-scan reusing the MCP index (refs Allow running mcp server with stdio transport #891, Facing slowness while running JSCPD  #591, Allow to provide custom store management engines. #583, @Jscpd/leveldb store does not support scanning multiple projects simultaneously #589)

Tier 2 — sustained asks or many third-party workarounds

• --history: duplication trend over git history, plus --ratchet to auto-tighten thresholds #1002 --history trend series and --ratchet (refs Walk back through the history and collect stats #77, Add hosted microservice for jscpd (tryjscpd) #355, Diff-aware detection and clone baseline: gate CI on new duplication only #944, OpenMetrics reporter #422)
• Skip import/using/require blocks and license-header comments by default #1003 Skip imports and license headers by default (refs Skip the import statements from the scan #415, Add a way to ignore copyright comments #788)
• Markdown: --no-prose (code-only) mode so documentation does not move code-duplication gates #1004 Markdown --no-prose (refs v5: Markdown (.md) files are never tokenized — the built-in markdown format matches 0 files #883)
• --ignore-add (additive ignore) and a warning when a run analyzes zero files via config #1005 --ignore-add and zero-files warning via config (refs fix(cli): unknown --format value silently scans 0 files instead of warning or failing; 'cs' also missing as csharp alias #964, Option to set code ignores in config file #215)
• Suppression report: list jscpd:ignore regions, their line totals and count #1006 Suppression report (refs Docs: --ignore-pattern and jscpd:ignore markers missing from v5 CLI reference #993, docs: document ignore patterns and inline markers #996)
• Config presets: --preset ai-strict (new-code gate <= 1%) and --preset tests-excluded #1007 Presets ai-strict, tests-excluded (refs Docs: Super Linter recipe — config file, baseline for pre-existing violations #994, Diff-aware detection and clone baseline: gate CI on new duplication only #944)
• Do not print tips/sponsor lines when stdout is not a TTY #1008 No tips when stdout is not a TTY (refs Add a way to ignore copyright comments #788, Optional first-run link for the AI reporter workflow #871)
• pr-summary reporter and a comment input for the GitHub Action (new duplication vs base branch) #1009 pr-summary reporter and Action comment input (refs Diff-aware detection and clone baseline: gate CI on new duplication only #944, Is jscpd support comparasions between commit diffs and all sources? #254, Comparing specific file in whole repo #588, Feature: Detect and fail on clones introduced by a Git diff #879)
• --lsp mode in the binary and an official VS Code extension #1010 --lsp and an official VS Code extension (refs VS code extension #571, Allow running mcp server with stdio transport #891)

Tier 3 — smaller or inferred demand, cheap wins, long-term bets

• Per-path and per-language overrides in .jscpd.json (minTokens, minLines, threshold, mode) #1011 Per-path / per-language overrides in config (refs Support isolated folder for monorepo #628, feat(rust): --skip-isolated folder groups for monorepos #942, Option to set code ignores in config file #215)
• MCP suggest_refactoring tool / refactor-brief reporter: LLM-ready brief per clone class #1012 MCP suggest_refactoring / refactor-brief reporter (refs Allow running mcp server with stdio transport #891, [rust] Folder/file summary report: top files by tokens, lines, size, and complexity #934, feat: add opt-in GCF output format for MCP tool responses (--gcf) #843, Add support for Codex for Skill install #800)
• Bitbucket Code Insights reporter and an Azure DevOps SARIF recipe #1013 Bitbucket Code Insights reporter, Azure DevOps recipe (refs Support GitLab / CodeClimate Code Quality Report format #958, OpenMetrics reporter #422, feat(sarif): expand rule information #914)
• Recipes for Nx, Turborepo, moon and a danger-js plugin #1014 Nx / Turborepo / moon recipes, danger-js plugin (refs Docs: Super Linter recipe — config file, baseline for pre-existing violations #994, Offer pre-commit Hook #493)
• Upstream: move Super Linter and MegaLinter off threshold 0 to --baseline-from-ref; fix the parallel report-path race #1015 Upstream Super Linter / MegaLinter defaults and report-path race (refs Docs: Super Linter recipe — config file, baseline for pre-existing violations #994; MegaLinter Unintentionally Enforces jscpd Threshold of 0 oxsecurity/megalinter#2450, #3979)
• Hosted duplication dashboard (trend history for many repositories) #1016 Hosted duplication dashboard (refs Add hosted microservice for jscpd (tryjscpd) #355, Walk back through the history and collect stats #77)
• Triage v4-era issues against the v5 engine and close what is obsolete or shipped #1017 Triage v4-era issues against v5 (refs Publish container to registry such as DockerHub/Github  #357, Facing slowness while running JSCPD  #591, Endless loop when using import way #731, Windows not showing report #645, the result is not accurate in the same file #590, jscpd exited with an error code ENOBUFS #519, Programming API exemples not working #416, Some specific .less lines take very long to check  #412, App takes long to end after work is done #411, Invalid XML format #375)

Positioning notes

• GitClear 2026: duplicated blocks per million changed lines up 81% since 2023; copy/paste share 9.4% (2022) → 15.7% (H1 2026) while moved/refactored code fell from 21% to 3.8%.
• HackerNoon 2026 ran jscpd over 49 vibe-coded repositories (7.2M lines): 25,612 clones, 7.98% average duplication, 4 clean repos.
• SonarSource (Mar 2026) recommends a ≤1% "duplicated lines on new code" gate for AI-generated code.
• jscpd npm downloads grew from 326k/month (Mar 2025) to 12.28M/month (Aug 2026); 241 CLAUDE.md files on GitHub reference jscpd.
• Agents ignore advisory output, so on-edit and CI hard gates (--watch and --cache: incremental re-scan on edit, reusing the MCP in-memory index #1001, Skip import/using/require blocks and license-header comments by default #1003, pr-summary reporter and a comment input for the GitHub Action (new duplication vs base branch) #1009) matter more than additional report formats.

Where jscpd already leads (not gaps)

Reporter breadth (SARIF + codeclimate + openmetrics + ai + xcode), built-in MCP and skills, git-ref baseline, 224 formats, sub-100 ms scans, blame comparison, monorepo isolation groups. PMD CPD still has no SARIF; dupFinder is sunset; the CodeClimate duplication engine is archived; SonarLint cannot show duplication in-editor.

Not verified

Reddit was unreachable during the research; DeepSource and CodeAnt AI duplication claims conflict across sources; whether v5 tokenizers already skip import blocks in practice was not tested.