feat: add two-layer duplicate code detection to PR check (#42) by kenjudy · Pull Request #46 · stride-nyc/code-quality-metrics

kenjudy · 2026-06-12T16:01:27Z

Summary

Adds Layer 1 static duplicate detection via jscpd (always-on, zero token cost): runs on every PR against changed production files, posts a ## Duplicate Code section in the PR comment when findings exist
Adds Layer 2 semantic duplicate detection via Claude (conditional on ANTHROPIC_API_KEY): scopes analysis to changed files plus their direct module neighbors, finds structural similarity that token-level matching misses
All four detection thresholds configurable in lib/config.js so teams using this on other repos (Java, Python, Go) can tune without touching library code

New files

lib/duplicate.js — runDuplicateCheck(filePaths) and resolveModuleNeighbors(filePaths)
__tests__/duplicate.test.js — 9 tests covering both functions
__tests__/config.test.js — 4 tests for new config key defaults

Changed files

lib/claude.js — adds analyzeDuplicatesWithClaude(client, filePaths, staticFindings)
lib/config.js — adds DUPLICATE_MIN_LINES, DUPLICATE_MIN_TOKENS, DUPLICATE_IGNORE_PATTERNS, DUPLICATE_SCAN_PATHS
.github/workflows/pr-metrics.yml — integrates both layers; PR comment shows static findings, semantic findings, and a layer indicator
CLAUDE.md — documents the four config keys; includes threshold tuning examples for Java, Python, and Go

Test plan

150 tests passing, 13 suites, no regressions
Coverage: 96.29% lines, 96.66% functions (thresholds: 80% / 90%)
Grep audit confirms both layer indicator strings present in YAML and Layer 2 guarded by ANTHROPIC_API_KEY check
runDuplicateCheck([]) returns [] without calling jscpd
resolveModuleNeighbors skips non-JS files, deleted files, and external imports
analyzeDuplicatesWithClaude(null, ...) returns [] without calling the API

🤖 Generated with Claude Code

Adds DUPLICATE_MIN_LINES (5), DUPLICATE_MIN_TOKENS (50), DUPLICATE_IGNORE_PATTERNS ([]), and DUPLICATE_SCAN_PATHS ([]) to CONFIG. All four keys are configurable so teams using this on other repos can tune thresholds without touching library code. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Adds Layer 1 static duplicate detection: runs jscpd via child_process, reads the JSON report it writes to a temp dir, and returns a structured array of {firstFile, secondFile, lines, tokens}. Returns empty array on empty input, missing report file, or non-zero jscpd exit — no throws. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Parses require()/import statements in JS/TS files, resolves relative paths to absolute, and returns the union of changed files plus their direct local imports. Non-JS files are passed through without parsing; files not on disk and external/node_modules imports are skipped. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Adds analyzeDuplicatesWithClaude(client, filePaths, staticFindings) which reads file contents, passes them to Claude with a duplication- focused system prompt, and returns structured findings with file1/file2/ similarity/concern/confidence. Degrades to [] when client is null or the API response is malformed. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Runs runDuplicateCheck against PR production files after checkout. Adds a 'Duplicate Code' section to the PR comment when findings exist, showing file:start-end pairs with line/token counts (capped at 10). Section is omitted silently when jscpd finds no duplicates. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Adds Layer 2 semantic duplicate detection when ANTHROPIC_API_KEY is set. Uses resolveModuleNeighbors to scope analysis to changed files plus their direct imports, then passes results to analyzeDuplicatesWithClaude. PR comment Duplication section now shows static findings, semantic findings (when found), and a layer indicator confirming which layers ran. Degrades to Layer 1 only when API key is absent. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…set.sh Uses PDCA_SKILL_DIR environment variable instead of an absolute path, with a guard that warns and skips the install if the directory is absent. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Adds four DUPLICATE_* keys to the configuration table with defaults and descriptions. Includes a tuning guide with example overrides for Java, Python, and Go repos where boilerplate inflates false positives. Updates the pr-metrics.yml architecture description to reflect the lib/duplicate.js and lib/claude.js dependencies added in this cycle. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

github-actions · 2026-06-12T16:01:47Z

PR Analysis

Size: large (based on production code)
Production Code: 299 lines (8 files)
Test Code: 178 lines (3 files)
Total: 477 lines (11 files)
Test-to-Production Ratio: 0.60:1

Concerns

Large production changes - review carefully
Median net additions ratio 0.93 exceeds 0.50 - possible batch-acceptance pattern

Strengths

Incremental development pattern detected
Includes refactoring or cleanup work
Message quality 100% meets discipline threshold

Commit Analysis

Total Commits: 9
Average Commit Size: 37 production lines
Average Files per Commit: 2.4

Metric	Value
Large commits (>100 prod lines)	0/9 (0%)
Sprawling commits (>5 files)	0/9 (0%)
Test-first discipline	4/9 (44%)
Message quality	9/9 (100%)
Median net additions ratio	0.93
Test-only commits	0
Production-only commits	5

Commit Details

c33afd6 Ken Judy (6/9/2026)
refactor: add duplicate detection config keys to lib/config.js
84 prod lines, 3 files [test+prod]

9bce469 Ken Judy (6/9/2026)
feat: implement runDuplicateCheck in lib/duplicate.js
50 prod lines, 3 files [test+prod]

ebc9063 Ken Judy (6/12/2026)
feat: implement resolveModuleNeighbors in lib/duplicate.js
33 prod lines, 4 files [test+prod]

bf1eac4 Ken Judy (6/12/2026)
feat: add semantic duplicate detection via lib/claude.js
58 prod lines, 3 files [test+prod]

13e2db9 Ken Judy (6/12/2026)
feat: integrate jscpd findings into pr-metrics.yml PR comment
19 prod lines, 2 files

23664a6 Ken Judy (6/12/2026)
feat: integrate semantic findings into pr-metrics.yml
40 prod lines, 2 files

d320a65 Ken Judy (6/12/2026)
chore: replace hardcoded PDCA skill path with env var in demo/temp-reset.sh
10 prod lines, 2 files

d601314 Ken Judy (6/12/2026)
docs: update CLAUDE.md with duplicate detection config parameters
27 prod lines, 1 files

d5d839c Ken Judy (6/12/2026)
chore: sync beads task state after PDCA cycle completion
16 prod lines, 2 files

Test Coverage

Test Adequacy: good

Good balance of production and test code

Target ratio: 0.5-2.0 test lines per production line

DORA Capability Assessment

Archetype: mixed-signals
No clear archetype pattern. Review individual metric thresholds.

Capability	Metric	Value	Target
Small Batches	Large commit %	0%	<20%
Small Batches	Sprawling commit %	0%	<10%
Version Control	Test-first discipline	44%	>50%
Version Control	Message quality	100%	>60%
AI Risk Signal	Net additions ratio (median)	0.93	<0.50

Duplicate Code

Found 1 duplicate block(s) across the changed files:

lib/claude.js:85-91 duplicates lib/claude.js:142-148 (7 lines, 64 tokens)
Layer 1 (static) ran. Set ANTHROPIC_API_KEY to enable semantic analysis.
Automated by Code Metrics Workflow

kenjudy and others added 9 commits June 9, 2026 11:15

chore: replace hardcoded PDCA skill path with env var in demo/temp-re…

d320a65

…set.sh Uses PDCA_SKILL_DIR environment variable instead of an absolute path, with a guard that warns and skips the install if the directory is absent. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

chore: sync beads task state after PDCA cycle completion

d5d839c

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

kenjudy linked an issue Jun 12, 2026 that may be closed by this pull request

Add two-layer duplicate code detection to PR check #42

Closed

kenjudy merged commit 986ee9f into main Jun 12, 2026
1 check passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: add two-layer duplicate code detection to PR check (#42)#46

feat: add two-layer duplicate code detection to PR check (#42)#46
kenjudy merged 9 commits into
mainfrom
42-add-two-layer-duplicate-code-detection-to-pr-check

kenjudy commented Jun 12, 2026

Uh oh!

github-actions Bot commented Jun 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

kenjudy commented Jun 12, 2026

Summary

New files

Changed files

Test plan

Uh oh!

github-actions Bot commented Jun 12, 2026

PR Analysis

Concerns

Strengths

Commit Analysis

Commit Details

Test Coverage

DORA Capability Assessment

Duplicate Code

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant