Content Engine

Content Engine (Xiaohongshu). Two modes: ① Deconstruct (v1) — input a viral XHS link, get an 18-field structured card. ② Generate (v2) — combine the deconstr...

Key things to check before installing or running this skill: 1) Clarify required credentials and endpoints. Ask the publisher which environment variables / API keys are required (TikHub/TikTok API, Ofox LLM keys, Nano Banana image keys, Volcengine/Seedance keys, etc.). The code references LLM/image/video clients but the skill metadata lists none — treat that as an under-declaration. Only provide keys you intend the skill to use, not broad cloud creds. 2) Inspect the code files that touch the network and filesystem first: scripts/content_engine/client.py, llm.py, nano_banana.py, seedance.py, generate.py, extract_xhs.py, and preflight.py. Search for strings like 'http', 'https', 'api', 'host', 'token', 'key', 'requests', 'urllib', 'socket'. Confirm which external hosts receive data and what data is transmitted (raw scraped notes, user-provided brand info, or full deconstruction outputs). 3) Run locally in a sandboxed environment. Execute preflight.py or run the scripts with a dry-run / test mode and with network egress blocked initially to see local behavior. Provide only minimal API keys in a restricted test account if you plan to enable network calls. Do not place unrelated secrets (AWS keys, GitHub tokens) into the agent environment. 4) Pay attention to data retention and privacy. The skill auto-writes deconstruction output and updates graph/ files that may contain scraped competitor content and comment text. If that is sensitive, store or encrypt graph/ files appropriately and review whether you want those artifacts written to persistent storage. 5) Confirm ffmpeg availability. SKILL.md says ffmpeg is required for video composition; install/allow only the ffmpeg binary and ensure paths are correct. The registry metadata should have listed this binary — request correction if it remains omitted. 6) Limit autonomous invocation if you are cautious. Autonomous execution is the platform default; if you want to reduce potential exposure until you audit the code, disable or restrict the skill's ability to run autonomously or run it only on-demand. 7) If you are not comfortable auditing the code yourself, ask the publisher for a short security/data-flow description: what endpoints are called, what data is sent, whether any logs or tokens are exfiltrated, and what minimal env vars are needed. Without that, treating this skill as 'runs code + networks + writes files' and running it in an isolated environment is the safest approach.

Type: OpenClaw Skill Name: qianxun-content-engine-en Version: 0.3.1 The skill bundle is a sophisticated content engine for Xiaohongshu that exhibits several high-risk capabilities. It performs network requests to multiple external APIs (TikHub, Ofox, Volcengine) and uses User-Agent spoofing (`curl/8.7.1` in `client.py`) to evade bot detection. It also executes shell commands via `subprocess` for media processing with `ffmpeg` (`video.py`, `video_compose.py`) and includes instructions in `SKILL.md` for the AI agent to autonomously modify its own knowledge graph files (`graph/` directory). While these behaviors are aligned with the stated purpose of content deconstruction and generation, the combination of network access, shell execution, and self-modifying file operations meets the threshold for a suspicious classification.