Back to skill
Skillv2.1.1
ClawScan security
Rag Architect · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
BenignMar 10, 2026, 2:29 PM
- Verdict
- benign
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill's code, documentation, and runtime instructions are coherent with a RAG design/optimization purpose; it asks for no credentials or installs and contains only local-processing Python scripts and guidance for RAG pipelines.
- Guidance
- This skill is coherent with its stated purpose and appears benign, but take these precautions before use: - Review and control which directory you give the scripts: the included code reads files recursively and will process any text/markdown files in the target path. Do not point it at sensitive directories unless you intend that. - The documentation recommends external services (OpenAI, Pinecone, etc.). Only provide API keys/credentials if you trust the integration and run the code in an environment you control. The skill itself does not request keys, so any credential sharing would be your action. - Run the scripts in an isolated or sandboxed environment (container or VM) if you want to avoid accidental access to other local data. - Inspect the full source (provided) before executing; although the visible code uses standard library file I/O and analysis, always verify there are no unexpected network calls or obfuscated code in omitted/truncated parts. - If you want stricter safety, disable autonomous invocation for this skill in your agent and call it manually after reviewing outputs and configuring credentials.
Review Dimensions
- Purpose & Capability
- okName and description (RAG Architect) match the provided SKILL.md and the three Python modules which implement chunking analysis, pipeline design, and evaluation. There are no unexplained environment variables, binaries, or external install steps that would be inconsistent with a RAG design tool.
- Instruction Scope
- noteSKILL.md is largely documentation and design guidance. It includes recommendations to read and analyze document corpora and to use production data (e.g., 'User Log Analysis' in evaluation). The included scripts (DocumentCorpus) explicitly read files from a directory — expected for this purpose but means the skill will need access to whatever directory you point it at. The guide also discusses integrating external services (OpenAI, Pinecone, Weaviate), which is normal for RAG design but implies you'll need to supply credentials if you follow those steps; the skill itself does not instruct to send data to any hidden endpoints.
- Install Mechanism
- okNo install specification is present and the Python files state they use only the standard library. Nothing is downloaded or written by an install step. This is a low-risk, instruction-and-code-only package.
- Credentials
- noteThe skill declares no required environment variables or primary credentials — proportional for a design/documentation tool. However, the documentation and code reference external embedding and vector DB services (OpenAI, Pinecone, Weaviate, etc.). Those integrations would require credentials if you choose to wire them up; the skill itself does not request or store secrets.
- Persistence & Privilege
- okalways is false and disable-model-invocation is false (normal). The skill does not request system-wide configuration changes or persistent privileges. Its runtime behavior is limited to local analysis of document directories when executed.