deep-research CLAUDE.md

deep-research CLAUDE.md is an instructions file for coding agents from treylom/deep-research. It costs 2,742 tokens per session, scanned B, original, MIT.

Repository instructions for deep-research, a project that provides a structured research process called SearchFlow. They cover installation, required tools, and how to copy the complete supporting files.

In plain words
What is it for?
Use them to install SearchFlow, verify the environment, and run research workflows with Claude Code or Codex CLI.
Why use it?
They prevent incomplete installation and explain how to check whether optional tools are available.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/treylom/deep-research/claude-md
Clone the repo
git clone --depth 1 https://github.com/treylom/deep-research

Wrote 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.

agentmods badge for deep-research CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/treylom/deep-research/claude-md.svg)](https://agentmods.dev/instructions/treylom/deep-research/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/treylom/deep-research/claude-md"><img src="https://agentmods.dev/badge/instructions/treylom/deep-research/claude-md.svg" alt="Measured on agentmods" height="20"></a>
Per session 2,742 This file is loaded in full into every session.
When invoked 2,742 The same file — it is already loaded in full.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.02742 $0.02742
Opus 5 $0.01371 $0.01371
Sonnet 5 $0.00548 $0.00548
Haiku 4.5 $0.00274 $0.00274

Measured today against content hash f28a7046fece, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

deep-research CLAUDE.md scanned grade B with 1 finding 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 today.

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.

Reads agent configuration directoriesmediumAgent snooping

.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.

**Codex CLI** — `~/.codex/config.toml` 에 **절대 경로**로 적습니다(플러그인 루트 변수는 없습니다):
CLAUDE.md · 168 lines

How it starts

The opening of the file, as written. The whole thing — 168 lines — stays where its author put it; the contents beside it link to each section on GitHub.

SearchFlow

리서치 공정. 진입점은 /searchflow 입니다. 개요는 README.md 를 보세요.

필요한 것

  • Claude Code 또는 Codex CLI (둘 중 하나)
  • node — 검사기 스크립트를 직접 실행합니다. 표준 내장 모듈만 쓰고 외부 패키지는 설치하지 않습니다(실측 환경 = v24.14.1)

설치

# 저장소 안에서 실행합니다.
# ⚠️ 폴더 이름을 그대로 주지 마세요 — `cp -r skills ~/.claude/skills/` 처럼 쓰면
#    대상 안에 `skills/skills/` 로 한 겹 더 들어갑니다. 끝에 `/*` 를 붙이면 안전합니다.
#    (macOS `/bin/cp` 실측: `skills/*`·`skills/` 는 안 들어가고, `skills` 만 들어갑니다.
#     GNU cp 는 미확인 — 확실한 `/*` 형태를 씁니다.)
mkdir -p ~/.claude/skills ~/.claude/commands ~/.claude/agents
cp -r skills/*   ~/.claude/skills/
cp -r commands/* ~/.claude/commands/
cp -r agents/*   ~/.claude/agents/

skills/searchflow/폴더 통째로 옮겨져야 합니다. 안에 scripts/(검사기)와 references/(공정 세부)가 들어 있고, 공정이 실행 중에 이 둘을 직접 읽습니다. SKILL.md 한 장만 복사하면 동작하지 않습니다.

설치 확인 (1줄)

node ~/.claude/skills/searchflow/scripts/env-detect.mjs

한 줄짜리 JSON 이 나오면 정상입니다. 예:

{"cross_engine":true,"knowledge_hook":false,"ooo":true,"multi_agent_api":"collab_v2"}

파일을 못 찾는다는 오류가 나면 scripts/ 가 같이 복사되지 않은 것입니다.

아무것도 없는 환경이면 이렇게 나오고, 그것도 정상입니다 — 저장소 밖 선택 도구가 없다는 뜻일 뿐 공정은 그대로 돕니다:

{"cross_engine":false,"knowledge_hook":false,"ooo":false,"multi_agent_api":"none"}

앞의 셋은 참/거짓이고 multi_agent_api 만 문자열(public·collab_v2·none)입니다 — 없을 때 값이 false 가 아니라 "none" 입니다.

종료 코드: 위처럼 인자 없이 실행하는 설치 확인은 항상 0 입니다(없음을 오류가 아니라 값으로 표현합니다). 다만 --test(자체 테스트)만은 실패 시 1 을 냅니다 — 이쪽은 "부재"가 아니라 "검사기가 깨졌다"는 뜻이라 0으로 덮으면 안 됩니다.

MCP 서버 (선택 — 공정을 하네스 밖에서 강제)

skills/searchflow/scripts/mcp-server.mjs 는 조사→제출→판정을 서버가 쥐는 stdio MCP 서버입니다. 없어도 스킬 단독으로 완주하고(2층 계약), 있으면 채점 기준이 조사 워커의 컨텍스트에 아예 안 실립니다.

Claude Code — 플러그인으로 설치하면 동봉된 .mcp.json 이 자동으로 잡습니다. 수동 등록은:

{ "mcpServers": { "searchflow": {
  "command": "node",
  "args": ["<이 저장소 경로>/skills/searchflow/scripts/mcp-server.mjs"] } } }

비대화형(claude -p)으로 돌릴 때는 도구가 보이는 것과 호출이 허용되는 것이 다릅니다. 등록만 하면 도구 목록에는 뜨지만 호출은 …but you haven't granted it yet 으로 막힙니다. 쓸 도구를 명시해 주세요 — 이름 앞에 서버 이름이 붙습니다:

Read the full file on GitHub · 168 lines

Changes

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.

  1. today Changed · -2 lines · +18 tokens per session f28a7046fece
  2. 4d ago First seen · 170 lines · 2,724 tokens per session scan B 4639c288997f

Subscribe to this mod's changes

deep-research CLAUDE.md is an instructions file published in the GitHub repository treylom/deep-research (2 stars, last pushed yesterday), licensed MIT. It adds 2,742 tokens to every session, about $0.0137 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.