Large Model Visual Question Answering Skill | 大模型视觉问答技能

Conducts open-ended Q&A on image content based on computer vision and large language models, supporting any questions to receive natural language responses....

This skill will upload images and metadata to external API endpoints (the code points to lifeemergence/open-api hosts by default) and will read/write config files and may create a local SQLite under the workspace. Before installing or running: 1) Confirm the remote API host and privacy policy — sensitive images should not be uploaded unless you trust the service. 2) Inspect or sandbox the skill (run in an isolated environment/container) — it can create files under OPENCLAW_WORKSPACE and save attachments. 3) Review workspace config files for secrets that the skill might read; avoid placing sensitive API keys or tokens in shared configs. 4) If you need strictly local-only VQA, do not use this skill. If anything is unclear, ask the publisher which endpoints receive images and what data is stored remotely vs locally.

Type: OpenClaw Skill Name: smyx-visual-qa-analysis Version: 1.0.0 The skill implements a complex integration with a third-party cloud service (lifeemergence.com) and contains highly controlling prompt instructions in SKILL.md that explicitly forbid the AI agent from accessing its own local memory or LanceDB, forcing it to rely solely on the remote API for history. The common library (smyx_common) includes logic in util.py and dao.py to automatically register/login users using identifiers like phone numbers, subsequently storing session tokens in a local SQLite database (smyx-common-claw.db). Additionally, AgentSkill in smyx_common/scripts/skill.py uses subprocess.run to recursively invoke the 'openclaw' agent, which is a high-privilege capability. While these features support the stated VQA and health analysis purposes, the aggressive override of agent memory and the automated credential management/persistence are high-risk behaviors.