feat(plugins): add verifying-phd-citations (APA 7 Korean + CrossRef DOI)

[dpyeye-commits]

What this PR adds

A focused Documentation plugin: verifying-phd-citations — verifies and formats APA 7 Korean academic citations across 14 source types, augments DOIs via CrossRef API, and applies CopyKiller 99-rule avoidance.

Why it fits this marketplace

The current Documentation category covers code/API docs but lacks academic-citation tooling. Korean academia (KCI journals, university dissertation requirements) imposes specific formatting rules on top of APA 7, and the dominant plagiarism checker (CopyKiller) requires citations follow exact patterns to avoid false-positive flags. This plugin fills that gap.

Files added

plugins/verifying-phd-citations/
├── .claude-plugin/
│   └── plugin.json          # plugin metadata
├── SKILL.md                 # Anthropic Skills format (gerund name, third-person English description with Korean trigger keywords)
├── run.mjs                  # executable Node.js — calls live CrossRef API, outputs APA 7 + in-text format
└── evals/
    └── evals.json           # 3 scenarios with expected_behavior arrays (Kahneman book, Korean thesis, et al. 3+)

Files modified

• .claude-plugin/marketplace.json — add plugin entry under plugins[]
• README.md — add link under ### Documentation section
• README-zh.md — same link in Chinese version

What it does

node plugins/verifying-phd-citations/run.mjs --query "Kahneman thinking fast slow 2011"
# Calls api.crossref.org, returns 5 candidates as APA 7 formatted strings + in-text + DOI

Smoke-tested against live CrossRef. The script explicitly flags candidates as review papers vs. original works so users do not blindly cite a review when they meant the original book — this is critical for academic integrity and is hardcoded into the skill instructions ("추측 금지" = no hallucination).

Network calls

Yes — calls https://api.crossref.org/works (CrossRef's public REST API, no auth required, polite User-Agent header included with maintainer email per CrossRef's etiquette guidelines). No other network calls. No analytics. No telemetry.

Permissions / safety

• No --dangerously-skip-permissions required
• No file writes outside the plugin directory
• No shell injection vectors (CrossRef query is URL-encoded)
• Safe to run with default Claude Code permissions

Alignment with Anthropic Skills best practices

Following docs.claude.com Agent Skills authoring guide:

• ✅ Gerund naming (verifying-phd-citations not phd-citation-verifier)
• ✅ Third-person English description with Korean trigger keywords for discovery
• ✅ SKILL.md under 500 lines (currently ~50)
• ✅ One-level reference depth
• ✅ 3 evaluation scenarios with expected_behavior
• ✅ No time-sensitive information
• ✅ Forward-slash paths only

Origin / maintenance

This plugin originated in my own marketplace at https://github.com/dpyeye-commits/my-arsenal-marketplace. I will keep both copies in sync. Happy to maintain it long-term.

Thanks for reviewing!