技能详情（站内镜像，无评论）
许可证：MIT-0
MIT-0 ·免费使用、修改和重新分发。无需归因。
版本：v2.1.3
统计：⭐ 3 · 2.3k · 9 current installs · 9 all-time installs
⭐ 3
安装量（当前） 9
🛡 VirusTotal ：良性 · OpenClaw ：可疑
Package：24601/agent-deep-research
安全扫描（ClawHub）
- VirusTotal ：良性
- OpenClaw ：可疑
OpenClaw 评估
The skill's code and instructions largely match its stated purpose (RAG-grounded research via Gemini), but there are small metadata inconsistencies and an important operational risk: the skill can upload local files to Google and will skip confirmations in non-interactive contexts — enabling an autonomous agent with filesystem access to exfiltrate data unless mitigated.
目的
Name/description describe deep research via Google Gemini; the code (research.py, upload.py, store.py, state.py) and required binary (uv) are consistent with that purpose. The skill legitimately needs a Google/Gemini API key and a runner like uv. Minor inconsistency: multiple env vars are declared in requires.env (including model/agent config vars) even though SKILL.md treats some as optional; this is a metadata overspec rather than functional…
说明范围
SKILL.md and scripts explicitly implement --context which uploads local files to Google ephemeral stores for RAG grounding. The instructions also state confirmations are auto-skipped when stdin is non-TTY (agent/CI mode). While the skill declares filtering of some sensitive filenames/extensions, that filter is finite and cannot guarantee all sensitive files are excluded. An autonomous agent with filesystem access could point --context at sensi…
安装机制
Install spec only references the 'uv' package runner (a known Python/astral tool). No downloads from arbitrary URLs or extract/install-from-unknown-host steps are present in the manifest. Code is included in the repo (no opaque binary blobs). This is a standard, low-risk install pattern for Python script-based skills.
证书
The skill requires a Google/Gemini API key (declares GEMINI_DEEP_RESEARCH_API_KEY, GOOGLE_API_KEY, GEMINI_API_KEY and picks first present) which is appropriate. Registry metadata also lists several model/agent config env vars (GEMINI_DEEP_RESEARCH_AGENT, GEMINI_DEEP_RESEARCH_MODEL, GEMINI_MODEL). SKILL.md treats those as optional, but the registry 'requires.env' lists all six — an over-declaration that can confuse operators. No unrelated cloud…
持久
always:false and default autonomous invocation are normal. The skill writes a local state file (.gemini-research.json) which (per docs) contains interaction IDs and hashes but not credentials. The bigger concern is operational: because confirmations are skipped in non-interactive mode and the skill can upload arbitrary files passed via --context, an autonomously-invoked agent with filesystem access can cause network uploads to Google. This com…
安装（复制给龙虾 AI）
将下方整段复制到龙虾中文库对话中，由龙虾按 SKILL.md 完成安装。
请把本段交给龙虾中文库（龙虾 AI）执行：为本机安装 OpenClaw 技能「Deep Research (Gemini)」。简介：Async deep research via Gemini Interactions API (no Gemini CLI dependency). RAG…。
请 fetch 以下地址读取 SKILL.md 并按文档完成安装：https://raw.githubusercontent.com/openclaw/skills/refs/heads/main/skills/24601/agent-deep-research/SKILL.md
（来源：yingzhi8.cn 技能库）SKILL.md
暂无本地缓存内容，可在后台执行详情同步。