← 返回 Skills 市场
etytabs

Reversal — Agent Input Reliability Layer

作者 Etytabs · GitHub ↗ · v1.0.2 · MIT-0
cross-platform ⚠ suspicious
36
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install reversal-engine
功能描述
Normalizes URLs, PDFs, Word, Excel, CSV, images, and text into clean JSON for reliable, universal input parsing in under 2 seconds.
使用说明 (SKILL.md)

Reversal — Agent Input Reliability Layer

Normalizes any source (URL, PDF, Word, Excel, CSV, image, dashboard screenshot) into clean structured JSON in under 2 seconds.

Stop your agents from hallucinating on raw HTML, failing on complex PDFs, or overflowing their context window on long documents. One call. Universal schema. Every format. Every time.

When to use this skill

  • Your agent needs to read a URL, article, or webpage without ads/noise
  • Your agent needs to parse a PDF, Word doc, or Excel file
  • Your agent needs to extract metrics from a dashboard screenshot
  • You have multiple sources and want a single normalized schema
  • You want deterministic, LLM-ready JSON instead of raw content

Tools

reverse_read(source)

Converts any source into structured JSON ready for LLM ingestion.

Input:  source  — URL (https://…) or file path (/path/to/file.pdf)
Output: { status, content_type, source, processed_in_ms, data: { title, summary_hint, word_count, content, … } }

Supported types: url · pdf · word · excel · csv · image · text

detect_content_type(source)

Lightweight probe — detects content type without full parse. Use before reverse_read to branch on type.

Input:  source  — URL or file path
Output: { source, detected_type }

batch_reverse(sources[])

Normalize up to 10 sources in a single call. Failed sources return an error field without aborting the batch.

Input:  sources  — array of 1–10 URLs or file paths
Output: [ { source, result } | { source, error } ]

upload_file(file_path)

Upload a local file; returns a file_id for use in reverse_read as file:\x3Cfile_id>. Validates extension before write. Supported: pdf, docx, xlsx, xls, csv, png, jpg, jpeg, webp, gif, txt, md.

Input:  file_path  — absolute path to local file
Output: { file_id, filename, size_bytes }

Install — stdio (local agents)

Works with Claude Desktop, Cursor, Windsurf, and any stdio MCP host.

git clone https://github.com/Etytabs/REVERSAL
cd REVERSAL
pip install -r requirements.txt
pip install -e .

Add to your MCP config:

{
  "mcpServers": {
    "reversal": {
      "command": "python",
      "args": ["-m", "reversal_engine.mcp_server"],
      "cwd": "/path/to/REVERSAL"
    }
  }
}

No API key required for URL, PDF, Word, Excel, CSV, and text parsing. Set ANTHROPIC_API_KEY only if you need image/dashboard screenshot parsing.

Install — HTTP (remote agents)

Works with OpenAI Codex, Claude Web, and any MCP-over-HTTP host.

# 1. Get a free API key
curl -X POST https://api.reversal.dev/v1/register \
  -H "Content-Type: application/json" \
  -d '{"email": "[email protected]"}'
# → { "api_key": "sk-rev-…" }
{
  "mcpServers": {
    "reversal": {
      "transport": "http",
      "url": "https://api.reversal.dev/v1/mcp",
      "headers": { "Authorization": "Bearer sk-rev-YOUR_KEY" }
    }
  }
}

For OpenAI Codex (~/.codex/config.toml):

[[mcp_servers]]
name = "reversal"
url  = "https://api.reversal.dev/v1/mcp"
[mcp_servers.headers]
Authorization = "Bearer sk-rev-YOUR_KEY"

Output schema

Every source returns the same envelope:

{
  "reversal_engine": "1.0",
  "status": "ok",
  "content_type": "url | pdf | word | excel | csv | image | text",
  "source": "original source",
  "processed_in_ms": 142,
  "data": {
    "title": "…",
    "summary_hint": "first 300 chars…",
    "word_count": 1240,
    "content": []
  }
}

Python SDK

pip install reversal-sdk
from reversal_sdk import ReversalClient

client = ReversalClient(api_key="sk-rev-…")

result = client.reverse("https://example.com/report")
print(result["summary"])

# Batch
results = client.batch([
    "https://example.com/report",
    "/path/to/data.xlsx",
])

# LangChain tool
from langchain.tools import tool

@tool
def reversal_tool(source: str) -> str:
    """Normalize any URL or file into structured JSON for the agent."""
    return client.reverse(source)["summary"]

TypeScript SDK

npm install reversal-client
import { ReversalClient } from "reversal-client";

const client = new ReversalClient({ apiKey: "sk-rev-…" });

const result = await client.reverse("https://example.com/report");
console.log(result.summary);

// Vercel AI SDK tool
import { tool } from "ai";
import { z } from "zod";

const reversalTool = tool({
  description: "Normalize any URL or file into structured JSON.",
  parameters: z.object({ source: z.string() }),
  execute: async ({ source }) => client.reverse(source),
});

Security

  • SSRF protection: all URLs validated against RFC-1918, loopback, link-local, and metadata endpoints
  • HMAC-SHA256 signed API keys — default secrets rejected at startup
  • JSON-only Redis cache — no pickle, no RCE surface
  • File uploads validated by extension allowlist before write
  • Optional OTP-gated registration (REVERSAL_REGISTER_REQUIRE_OTP=true)
  • Per-IP rate limiting on registration (3/min)
  • Non-root Docker runtime (uid 1001)
  • No obfuscated install commands · No curl|sh · No undeclared secrets

Repository

https://github.com/Etytabs/REVERSAL

安全使用建议
Install only if you trust the external Reversal repository and service. Prefer the local setup for sensitive documents, keep API keys protected, and ask the agent to parse only specific files or URLs you intend to share.
功能分析
Type: OpenClaw Skill Name: reversal-engine Version: 1.0.2 The `SKILL.md` file defines tools that allow an agent to read arbitrary URLs and local files via absolute paths (`upload_file`), which are then processed or transmitted to an external endpoint (`api.reversal.dev`). While the documentation describes security mitigations like SSRF protection and extension allowlisting, the inherent capability to access sensitive local data and send it to a third-party service poses a significant risk of data exfiltration if the agent is misdirected. No evidence of intentional malice, such as obfuscation or hidden prompt injections, was found in the provided files.
能力标签
requires-sensitive-credentials
能力评估
Purpose & Capability
The stated purpose—normalizing URLs, documents, images, and text into JSON—is coherent with the described tools, but the reviewed package contains only SKILL.md and delegates actual functionality to external MCP/API/SDK components.
Instruction Scope
The instructions let the agent process URLs, absolute local file paths, and batches of sources. This is purpose-aligned, but users should keep file access explicit.
Install Mechanism
There is no registry install spec or included code; the skill provides manual GitHub, pip, npm, and MCP configuration steps. These are disclosed but not reviewed in this artifact set.
Credentials
Remote Reversal API use and optional Anthropic image parsing are proportional to the parsing purpose, but they may send selected content outside the local environment.
Persistence & Privilege
No hidden background persistence or privilege escalation is shown, but the skill uses API keys and mentions Redis caching, so users should check retention and credential handling.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install reversal-engine
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /reversal-engine 触发
  4. 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
v1.0.2
Reversal-engine 1.0.2 changelog: - Added comprehensive SKILL.md with clear usage guide, API details, and supported formats. - Introduced four main tools: reverse_read, detect_content_type, batch_reverse, and upload_file. - Unified output schema for all source types with deterministic, LLM-ready JSON. - Outlined installation steps for both local (stdio) and remote (HTTP) agent environments. - Published security guarantees, Python and TypeScript SDK setup, and multi-agent compatibility.
元数据
Slug reversal-engine
版本 1.0.2
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Reversal — Agent Input Reliability Layer 是什么?

Normalizes URLs, PDFs, Word, Excel, CSV, images, and text into clean JSON for reliable, universal input parsing in under 2 seconds. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 36 次。

如何安装 Reversal — Agent Input Reliability Layer?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install reversal-engine」即可一键安装,无需额外配置。

Reversal — Agent Input Reliability Layer 是免费的吗?

是的,Reversal — Agent Input Reliability Layer 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Reversal — Agent Input Reliability Layer 支持哪些平台?

Reversal — Agent Input Reliability Layer 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 Reversal — Agent Input Reliability Layer?

由 Etytabs(@etytabs)开发并维护,当前版本 v1.0.2。

💬 留言讨论