Skip to content

Latest commit

 

History

History
121 lines (88 loc) · 4.44 KB

CHANGELOG-draft.md

File metadata and controls

121 lines (88 loc) · 4.44 KB

AgentLinter Changelog

v2.1.0 — 2026-03-XX 🚀

🆕 New: 13 Rules Added (17 → 30 total)

📊 Token Bloat Score 강화 (+3 rules)

  • clarity/duplicate-content — 파일 내/파일 간 중복 섹션 탐지 (3-gram Jaccard similarity ≥ 0.6)
  • clarity/obvious-statements — LLM에게 불필요한 "당연한 말" 감지 ("be accurate", "follow instructions" 등)
  • clarity/token-budget-range — 실제 토큰 수 추정 기반 그레이딩 (English: 4 chars/token, Korean: 1.5 chars/token)

🕐 Freshness / Staleness Detector (+3 rules, NEW)

  • consistency/stale-file-reference — Markdown 내 파일 경로가 실제로 존재하는지 검증
  • consistency/stale-date — 90일+ 이전 날짜 경고, 180일+ error
  • consistency/stale-package-reference — 참조된 패키지가 package.json에 있는지 크로스체크

📎 @ Import Validator (+2 rules, NEW)

  • structure/dead-import — @file.md 스타일 import가 실제 파일을 가리키는지 검증
  • structure/circular-import — import 그래프에서 순환 참조 탐지

🔀 Multi-Framework Support (+4 rules, NEW)

  • Cursor (.cursorrules) 스캔 지원 + 형식 검증
    • cursor/rules-format — .cursorrules 구조 검증
    • cursor/no-conflicting-claude — CLAUDE.md와 충돌 감지
  • GitHub Copilot (.github/copilot-instructions.md) 스캔 지원
    • copilot/instructions-format — 형식 검증
    • copilot/no-conflicting-claude — CLAUDE.md와 충돌 감지

✍️ Describe vs Command 분류기 (+1 rule)

  • claude-code/descriptive-ratio — 서술문 vs 명령문 비율 분석, descriptive > 60%이면 경고

🔬 Enhanced

  • Token Efficiency Score: line-based → token-estimated grading (Korean 보정 포함)
  • Parser: .cursorrules, .github/ 디렉토리 스캔 추가
  • LintContext type 확장: cursor, copilot context 추가

📚 Research-backed

  • 웹사이트에 "Research-backed linting" 배너 추가
  • Gloaguen et al. (2026) 레퍼런스

v2.0.0 — 2026-03-04 🚀

Major: v2 Analysis Engine

🧠 v2 Analyzers (5 modules)

  • Cognitive Load — measures instruction density and mental overhead
  • Token Heatmap — visualizes token distribution across file sections
  • Modularity — evaluates separation of concerns and file organization
  • Role Complexity — detects over-complex role/persona definitions
  • Security Scan — deep security analysis with 25 patterns (up from 15)

🎯 Clarity Score

  • 17 ambiguous pattern detections (Korean + English)
  • Weighted scoring with rewrite suggestions
  • Korean token correction for accurate estimation

💡 Actionable Suggestions

  • Every issue tagged with priority: HIGH / MED / LOW
  • Fix hints with concrete rewrite examples

🏷️ Badge API

  • /api/badge?score=N endpoint — embed SVG score badge in README

🔐 Security Patterns Expanded

  • 25 patterns total (AWS, JWT, injection, role-hijacking, etc.)

🌏 Korean Token Correction

  • Accurate token estimation for Korean agent files

🖥️ Token Map UI

  • Budget gauge visualization
  • .claudeignore card display
  • v2 analyzer integration in web UI

v1.2.0 — 2026-03-03

New Features

  • 📦 Token Budget Estimator — per-file token usage calculation with budget gauge
  • 🚫 .claudeignore Rules — exclude files from context window analysis

v1.1.0 — 2026-03-01

New Features

  • 🎯 Position Risk Warning (structure/position-risk-warning)
  • 📊 Token Efficiency Score (clarity/token-efficiency-score)
  • 🔐 Enhanced Security Check — prompt injection + API key exposure detection

v1.0.0 — 2026-02-25 🎉

ESLint for AI Agents — first stable release

Features

  • 17 lint rules across 9 categories
  • Claude Code + OpenClaw runtime support
  • CLI (agentlinter .) + Web UI
  • Instruction count, relevance trap, context budget estimator
  • .claude/ full directory scan
  • Skills safety audit

Pre-1.0

  • v0.9.0 — Token Budget Checker, Instruction Scope, Skills Security+, Hooks Advisor
  • v0.8.2 — 4 new rules + Remote-Ready Score category
  • v0.8.1 — False positive bug fixes (5)
  • v0.8.0 — 7 new rules + Claude Code latest spec
  • v0.7.1 — Runtime scoring improvement (55→80) + tone detection fix
  • v0.7.0 — Comprehensive best practices upgrade
  • v0.6.1 — RFC 2119, file reference, security skill context fixes
  • v0.6.0 — Stricter english-config-files threshold, clarity rule