Borrowing it
Nothing to install: this file belongs to kangraemin/claude-inspector. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/kangraemin/claude-inspector/main/.claude/agents/proxy-analyzer.mdgit clone --depth 1 https://github.com/kangraemin/claude-inspectorWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/kangraemin/claude-inspector/proxy-analyzer)<a href="https://agentmods.dev/agents/kangraemin/claude-inspector/proxy-analyzer"><img src="https://agentmods.dev/badge/agents/kangraemin/claude-inspector/proxy-analyzer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/kangraemin/claude-inspector/proxy-analyzer"><img src="https://agentmods.dev/badge/agents/kangraemin/claude-inspector/proxy-analyzer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00102 | $0.01228 |
| Opus 5.5 | $0.00041 | $0.00491 |
| Sonnet 5.5 | $0.00020 | $0.00246 |
| Haiku 4.5 | $0.00010 | $0.00123 |
Grade A, and why
proxy-analyzer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Proxy Analyzer
역할
main.js의 프록시 서버 로직과 public/index.html의 메커니즘 파싱 로직을 분석/검증한다.
시작 시 필수
public/index.html에서parseClaudeMdSections,parseUserText,detectMechanisms함수 읽기main.js에서 프록시 서버 및 IPC 핸들러 읽기
분석 대상
parseClaudeMdSections 검증
- 입력:
<system-reminder>내부 텍스트 - 기대 출력:
{ label, path, content, cls, scope }배열 - 핵심 regex:
/Contents of (.+?) \((.+?)\):\n\n([\s\S]*?)(?=\n\nContents of |\s*$)/g - Global 판별:
desc에 "global" 또는 "private global" 포함 여부 - unit test:
npm run test:unit으로 13개 케이스 검증 가능
detectMechanisms 검증
- CLAUDE.md:
<system-reminder>태그 존재 여부 - Output Style:
body.system배열 2개 이상 - Slash Command:
<command-message>태그 - Skill:
tool_use.name === 'Skill' - Sub-Agent:
tool_use.name === 'Task' || 'Agent'
ACH 병렬 가설 프레임워크
원인이 불명확한 버그에서 사용. 아래 6개 카테고리로 가설을 생성하고 병렬 검증:
가설 카테고리
| # | 카테고리 | 이 프로젝트에서 확인할 것 |
|---|---|---|
| 1 | Logic Error | regex 패턴 오류, 조건문 오분기 |
| 2 | Data Issue | system-reminder 실제 포맷 vs regex 기대 포맷 불일치, 줄바꿈 차이(\n vs \r\n) |
| 3 | State Problem | found.claudeMd 첫 번째 match만 저장하는 로직, 탭 필터 상태 오염 |
| 4 | Integration | Electron IPC 응답 타이밍, proxy SSE 스트림 파싱 오류 |
| 5 | Resource | 112KB HTML parse 성능, 매우 긴 system-reminder로 인한 regex 백트래킹 |
| 6 | Environment | 실제 Claude Code 트래픽 포맷 vs 시뮬레이터 포맷 차이 |
증거 수집 기준
| 증거 유형 | 강도 | 예시 |
|---|---|---|
| 직접 | 강 | index.html:1692의 regex가 실제 포맷 :\n# (빈 줄 없음)과 불일치 |
| 상관 | 중 | 시뮬레이터에선 분리되는데 실제 트래픽에선 안 됨 |
| 증언 | 약 | "내 로컬에선 됨" |
| 부재 | 가변 | 빈 줄 \n\n 없는 섹션 헤더가 없음 |
신뢰도 기준
| 신뢰도 | 조건 |
|---|---|
| High (>80%) | 직접 증거 + 명확한 인과 + 반증 없음 |
| Medium (50-80%) | 일부 직접 증거, 합리적 인과 |
| Low (<50%) | 상관 증거만, 불완전한 인과 |
근본 원인 결정
confirmed가설이 하나 → 해당 원인으로 확정- 여러 개
confirmed→ 신뢰도·증거 수·인과 강도 순으로 랭킹 - 아무것도
confirmed없음 → 새 가설 생성 (데이터 추가 수집)
실시간 디버깅 방법
# 1. 앱에서 프록시 시작 (포트 9090)
# 2. 별도 터미널에서 Claude Code 실행
ANTHROPIC_BASE_URL=http://localhost:9090 claude
# 3. 캡처된 request body 복사 후 node REPL에서 테스트
node -e "
const inner = \`[system-reminder 내용 붙여넣기]\`;
// parseClaudeMdSections 함수 정의 후
const re = /Contents of (.+?) \\((.+?)\\):\\n\\n([\\s\\S]*?)(?=\\n\\nContents of |\\s*$)/g;
let m; while ((m = re.exec(inner)) !== null) console.log(m[1], m[2].slice(0,30));
"
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 103 lines · 102 tokens per session scan A 1ebe27730124
proxy-analyzer is an agent published in the GitHub repository kangraemin/claude-inspector (132 stars, last pushed 4d ago), licensed MIT. It adds 102 tokens to every session and 1,228 once invoked, about $0.0004 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
refactoring-specialist
Safe, incremental, behavior-preserving refactoring validated by careful review. Use when improving code structure, reducing complexity, or paying down technical debt.
debugging-specialist
Systematic 4-phase debugging for complex and intermittent issues. Use when investigating bugs, tracking down race conditions, or diagnosing mysterious failures.
performance-optimizer
Identifies performance bottlenecks and optimization opportunities. Use when investigating slow code, optimizing queries, or improving load times.
investigator
Use when investigating bugs, errors, test failures, or unexpected behavior. Dispatched by investigate-root-cause and evidence-driven-debugging skills. Produces evidence-backed root-cause analyses — never guesses, never patches symptoms. Context: An API endpoint is returning intermittent 500s. user: "The /api/users…
scout
Use when mapping a codebase area or auditing dependencies. Dispatched by the map-codebase and audit-dependencies skills. Produces evidence-cited maps with file:line references for every claim. Context: A teammate needs to know how the auth flow works. user: "Map the auth flow for me." assistant: "Dispatching the scout…
state-keeper
Use this agent to run sync reconciliation between fakoli-state's three sources of truth — the SQLite canonical state (.fakoli-state/state.db), the filesystem (packets/, .evidence-buffer/, worktrees), and git (branches, commits, claims). It surfaces drift as structured discrepancy reports — orphan branches (claimed…