← 返回 Skills 市场
LLMWhisperer
作者
Gustavo Madeira Santana
· GitHub ↗
· v0.0.7
3010
总下载
1
收藏
6
当前安装
7
版本数
在 OpenClaw 中安装
/install llmwhisperer
功能描述
Extract text and layout from images and PDFs using LLMWhisperer API. Good for handwriting and complex forms.
使用说明 (SKILL.md)
LLMWhisperer
Extract text from images and PDFs using the LLMWhisperer API — great for handwriting and complex forms.
Configuration
Requires LLMWHISPERER_API_KEY in ~/.clawdbot/.env:
echo "LLMWHISPERER_API_KEY=your_key_here" >> ~/.clawdbot/.env
Get an API Key
Get a free API key at unstract.com/llmwhisperer.
- Free Tier: 100 pages/day
Usage
llmwhisperer \x3Cfile>
Script Source
The executable script is located at scripts/llmwhisperer.
#!/bin/bash
# Extract text using LLMWhisperer API
if [ -z "$LLMWHISPERER_API_KEY" ]; then
if [ -f ~/.clawdbot/.env ]; then
# shellcheck disable=SC2046
export $(grep -v '^#' ~/.clawdbot/.env | grep 'LLMWHISPERER_API_KEY' | xargs)
fi
fi
if [ -z "$LLMWHISPERER_API_KEY" ]; then
echo "Error: LLMWHISPERER_API_KEY not found in env or ~/.clawdbot/.env"
exit 1
fi
FILE="$1"
if [ -z "$FILE" ]; then
echo "Usage: $0 \x3Cfile>"
exit 1
fi
curl -s -X POST "https://llmwhisperer-api.us-central.unstract.com/api/v2/whisper?mode=high_quality&output_mode=layout_preserving" \
-H "Content-Type: application/octet-stream" \
-H "unstract-key: $LLMWHISPERER_API_KEY" \
--data-binary "@$FILE"
Examples
Print text to terminal:
llmwhisperer flyer.jpg
Save output to a text file:
llmwhisperer invoice.pdf > invoice.txt
Process a handwritten note:
llmwhisperer notes.jpg
安全使用建议
This skill appears to do what it says (send a file to the LLMWhisperer API to extract text/layout), but there are some inconsistencies you should address before installing or using it with sensitive data:
- Manifest mismatch: The registry entry claims no required env vars or config paths, but SKILL.md requires LLMWHISPERER_API_KEY stored in ~/.clawdbot/.env. Ask the publisher to update the manifest to declare the required env var and config path.
- Missing script file: SKILL.md references scripts/llmwhisperer but the package contains only SKILL.md. Confirm how the script will be installed or executed by your platform.
- Data privacy: Using the skill uploads your file to a third‑party API (llmwhisperer-api.us-central.unstract.com). Do not upload sensitive or regulated documents unless you trust the service and have reviewed its privacy/security policy and TLS/certificate validity.
- Minor robustness: The inline export-loading command is brittle (word-splitting) but not obviously malicious. If you use the skill, prefer setting LLMWHISPERER_API_KEY in your environment rather than relying on parsing a shared .env file.
If you cannot verify the vendor (unstract) or do not trust sending documents off-host, consider running an offline OCR alternative instead. If you want, I can draft a message to the skill owner asking them to fix the manifest and include the script file.
功能分析
Type: OpenClaw Skill
Name: llmwhisperer
Version: 0.0.7
The skill is benign. It reads a specific API key (`LLMWHISPERER_API_KEY`) from `~/.clawdbot/.env` (a standard practice for OpenClaw skills) and uses it to authenticate a `curl` request. The `curl` command sends the user-provided file to the legitimate LLMWhisperer API endpoint (`https://llmwhisperer-api.us-central.unstract.com`) for text extraction, aligning perfectly with the skill's stated purpose. There is no evidence of data exfiltration beyond the necessary API key and input file, malicious execution, persistence mechanisms, or prompt injection attempts in SKILL.md.
能力评估
Purpose & Capability
The SKILL.md and the inline script perform exactly what the description says: POST the provided file to an LLMWhisperer OCR endpoint and return the result. The external endpoints in the script (unstract.com / llmwhisperer-api.us-central.unstract.com) match the documentation links in the README. However the skill's registry metadata lacks a homepage/source URL despite the SKILL.md referencing unstract — that's a minor provenance gap.
Instruction Scope
Instructions are focused: they read an API key from ~/.clawdbot/.env (or the environment) and upload the specified file. The script does not attempt to read other system paths or other credentials. Two points to note: (1) the SKILL.md claims an executable at scripts/llmwhisperer but no such file exists in the package (the script is only embedded in the doc), and (2) the script uses export $(grep ... | xargs) to load the key, which is a common pattern but can mis-handle unusual .env lines (word-splitting) — not an obvious exfiltration vector, but brittle.
Install Mechanism
There is no install spec (instruction-only), so nothing is written to disk by a package installer. This is lower risk. The only install-like action is the README telling the user how to append the API key to ~/.clawdbot/.env. The missing scripts/ path noted above is an inconsistency but not an installer risk by itself.
Credentials
The skill requires a single service credential (LLMWHISPERER_API_KEY) which is appropriate for a third‑party OCR API. However the registry metadata declares no required env vars or config paths while the SKILL.md explicitly requires LLMWHISPERER_API_KEY and the ~/.clawdbot/.env file — this mismatch is an incoherence in the manifest and should be corrected. Also note: the skill will upload any file you pass to an external, third‑party API; that is expected for this purpose but is a privacy/security consideration for sensitive documents.
Persistence & Privilege
The skill is not marked always:true and does not request persistent system-level privileges. It is user-invocable and can be invoked autonomously (default), which is normal for skills; there are no other privilege escalation indicators in the manifest or instructions.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install llmwhisperer - 安装完成后，直接呼叫该 Skill 的名称或使用
/llmwhisperer触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v0.0.7
- Added the full llmwhisperer shell script directly to SKILL.md for easier reference.
- Clarified script location as scripts/llmwhisperer instead of a user-home subdirectory path.
- No changes to installation or usage instructions; example usage unchanged.
v0.0.6
Fix: include scripts folder
v0.0.5
Include scripts folder
v0.0.4
Add scripts folder with executable for easy installation
v0.0.3
Added usage examples to docs
v0.0.2
Moved script to root to fix missing file
v0.0.1
Initial release
元数据
常见问题
LLMWhisperer 是什么？
Extract text and layout from images and PDFs using LLMWhisperer API. Good for handwriting and complex forms. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 3010 次。
如何安装 LLMWhisperer？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install llmwhisperer」即可一键安装，无需额外配置。
LLMWhisperer 是免费的吗？
是的，LLMWhisperer 完全免费（开源免费），可自由下载、安装和使用。
LLMWhisperer 支持哪些平台？
LLMWhisperer 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 LLMWhisperer？
由 Gustavo Madeira Santana（@gumadeiras）开发并维护，当前版本 v0.0.7。
推荐 Skills