技能详情（站内镜像，无评论）
作者：pollo @arespollo
许可证：MIT-0
MIT-0 ·免费使用、修改和重新分发。无需归因。
版本：v1.0.0
统计：⭐ 0 · 536 · 5 current installs · 6 all-time installs
⭐ 0
安装量（当前） 6
🛡 VirusTotal ：可疑 · OpenClaw ：可疑
Package：arespollo/grok-research
安全扫描（ClawHub）
- VirusTotal ：可疑
- OpenClaw ：可疑
OpenClaw 评估
The skill mostly does what it says (forwards queries to a Grok-like API) but has a few inconsistencies and provenance/installation gaps you should understand before installing — and it will forward user text to an external service.
目的
The skill's code and SKILL.md align with the stated purpose: it forwards user queries to a remote Grok API and returns the model output. However the description claims 'real-time X/Twitter knowledge' (an external capability of the Grok service) but the skill itself does not access X/Twitter — it only proxies to the ai.a9.bot endpoint. This claim therefore depends entirely on the remote service, not on the skill. Also the top-of-file comment re…
说明范围
SKILL.md instructs the agent to forward the user's message 'as-is' to the Grok API and the code does exactly that. That means any prompt-injection content in user input will be forwarded to the external API — the marketing claim 'no prompt injection' is misleading: the skill avoids adding prompts, but does not sanitize or block injected content. The SKILL.md also suggests running the script with bun, but the declared required binaries list is …
安装机制
There is no explicit install spec (instruction-only), which reduces risk, but the script requires the 'bun' runtime (shebang and usage examples). The registry metadata lists no required binaries — that is an inconsistency and may lead to runtime surprises. No external downloads or archives are used.
证书
Only a single API credential (A9_GROK_API_KEY) is required, which is proportionate to the stated purpose. Caveat: the source owner and homepage are unknown and the destination host (https://ai.a9.bot) is not documented in registry metadata; supplying your API key will send user queries to that third-party endpoint.
持久
The skill does not request persistent installation privileges (always:false), does not modify other skills or system settings, and contains no install hooks. It runs as a simple proxy CLI when invoked.
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「Grok Research」。简介：Crypto research via Grok model's real-time X/Twitter knowledge. Forwards the us…。
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/arespollo/grok-research/SKILL.md
（来源：yingzhi8.cn 技能库）SKILL.md
---
name: grok-research
description: >
Crypto research via Grok model's real-time X/Twitter knowledge.
Forwards the user's query as-is to Grok API — no prompt injection, no context bloat.
Use when: (1) user asks to research a token's narrative/story/sentiment,
(2) user says "调研", "research", "grok research", "查一下叙事", "帮我看看这个币",
(3) user wants to know what CT is saying about a token/project.
NOT for: price analysis, on-chain data, or trading execution.
---
# Grok Research
Forward user's research query directly to Grok API. No extra prompts — what the user says is exactly what Grok receives.
## Config
Env var `A9_GROK_API_KEY` required. API base URL: `https://ai.a9.bot/v1` (hardcoded).
## Usage
```bash
cd ~/.openclaw/workspace/skills/grok-research
bun run grok-research.ts <query>
bun run grok-research.ts --model grok-4.20-beta <query>
```
Default model: `grok-4.20-beta`. Another model: `--model grok-4.1-thinking`.
## How to Call
Pass the user's original message as the query. Do not add system prompts or templates — forward as-is.
Example: user says "调研一下代币叙事 $buttcoin Cm6fNnMk..."
→ `bun run grok-research.ts "调研一下代币叙事 $buttcoin Cm6fNnMk..."`
## Output
- stdout: Grok's response (forward to user)
- stderr: status/errors
- Format for Discord before sending (no markdown tables)