OEE Knowledge Base RAG

Semantic knowledge base allowing ingest, search, and retrieval of saved texts, URLs, and files using embeddings and SQLite.

This skill appears to be a local RAG knowledge base and the code largely matches that purpose, but there are important inconsistencies you should resolve before installing or supplying API keys: - SKILL.md says you need ANTHROPIC_API_KEY or OPENAI_API_KEY but the registry metadata lists no required env vars — assume the code will attempt to call an embeddings API. Only provide a key you trust and ideally create a dedicated, limited-scope key for this purpose. - The code calls external binaries (yt-dlp for YouTube transcripts, pdftotext as a PDF fallback) and may rely on Python packages like PyMuPDF (fitz). These dependencies are not declared; if you run this skill, ensure those binaries/packages are installed from official sources. Running in a sandbox/container is recommended. - The skill fetches arbitrary URLs and runs subprocesses to extract content. This is expected for an ingest tool but means it will make outbound network requests and execute local binaries. Do not point it at sensitive internal URLs or supply credentials without reviewing the code. - The skill creates a local database (kb.db) in the skill directory and writes temporary files to /tmp; protect that file if it will contain sensitive data. If you plan to use this skill: inspect the full kb.py (the provided snippet was truncated), confirm where it sends network requests and which env vars it reads, install required binaries from trusted sources, and consider running it in an isolated environment. If you cannot verify those points, treat the skill as potentially risky.

Type: OpenClaw Skill Name: oee-knowledge-base Version: 1.0.0 The `kb.py` file contains a critical shell injection vulnerability within the `extract_pdf` function. The `subprocess.run` call, which executes a Python script via `python3 -c`, constructs the script using an f-string that directly embeds the `path` variable (derived from user-provided `url_or_path`) without proper sanitization. A malicious URL or local path containing shell metacharacters or Python string delimiters could lead to arbitrary command execution on the host system. Additionally, the skill utilizes `subprocess.run` to execute external tools like `yt-dlp` and `pdftotext`, which, while intended, expands the attack surface.