RAG Search

Backend retrieval skill for structured search of occupational health standards and documents, returning relevant text with source and clause details.

Security Scan

The skill's code generally matches a local RAG retrieval purpose, but it accesses hard-coded local paths and external embedding/rerank clients without declaring required config paths or credentials, which is an incoherence you should verify before installing.

Purpose & Capability

Name, SKILL.md and handler.py consistently implement a minimal RAG search against a local vector DB for occupational-health regulations. However, the code requires a specific workspace layout (/root/.openclaw/workspace/rag_system/...) and a vectors.db file that are not declared in the skill metadata (no required config paths). That mismatch (implicit dependency on a local repo) is unexpected and should be justified.

Instruction Scope

SKILL.md describes only querying and returning original text. The implementation imports modules from an absolute path and reads /root/.openclaw/.../data/vectors.db. The handler returns raw document content from the DB (potentially sensitive). SKILL.md also warns this is a backend-only component, but the skill metadata does not mark any special protection; the instructions/code will access local files outside the skill bundle at runtime.

Install Mechanism

There is no install spec and no remote downloads; the skill is instruction-only aside from a local handler.py. This minimizes installation risk because nothing is fetched or written by an install step.

Credentials

handler.py instantiates QwenEmbeddingClient and QwenRerankClient which likely require API credentials or network access, but the skill declares no required environment variables or primary credential. Additionally, it relies on a hard-coded filesystem path rather than a declared config path. Missing declarations (API keys, DB path) are incoherent and could hide secret requirements or unexpected network calls.

Persistence & Privilege

The skill does not request always:true and does not modify system configs. However it alters sys.path at runtime to import code from /root/.openclaw/workspace, granting it access to other code and data in that workspace—this pattern increases its blast radius compared with a self-contained skill.

What to consider before installing

The code implements a local RAG query but relies on an absolute workspace path and external embedding/rerank clients while declaring no config paths or credentials. Before installing: (1) confirm the runtime will host /root/.openclaw/workspace/rag_system/data/vectors.db and that returning raw DB content is acceptable; (2) ask the author for the implementations of embedding_client and search_pipeline (they may require API keys or network access); (3) require the skill manifest to declare any required config paths and env vars (e.g., API keys, DB location); (4) review embedding/rerank client code for outbound network calls or credential usage to avoid secret exfiltration; (5) avoid exposing this skill directly to end users until these questions are resolved.

Minimal RAG search skill for backend retrieval.

⚠️ Important

This skill is intended to be used as a backend retrieval component and should not be invoked directly by end users.

Use occupational_health_qa or occupational_health_report_writer for direct user requests.

Usage

你：调用 rag-search，查询"GBZ 2.1-2019 苯 职业接触限值"

Returns

Returns structured search results with:

• content: Original text from the document
• source: File name / standard number
• clause: Clause number (if available)
• regulation_level: Regulation level (国家法律/国家标准/行业标准/etc)
• score: Relevance score (0-1)

Example Response

{
  "results": [
    {
      "content": "苯的时间加权平均容许浓度（PC-TWA）为6 mg/m³...",
      "source": "GBZ 2.1-2019.pdf",
      "clause": "第4.1条",
      "regulation_level": "国家标准",
      "score": 0.93
    }
  ]
}