半饱 — 生活的高潮所在。A mindful eating companion for desk workers. Track meals with photos, understand your body's needs, no gym guilt.

What to consider before installing: - Data storage and privacy: the skill stores meal logs, weight, and activity locally (data/*.json). It may save file paths to photos from your machine in logs; although the scripts do not upload photos, those paths can appear in stored records and responses. Review where the skill will be installed so you control the data directory. - Silent health inferences: AGENT_GUIDE documents that the agent should silently infer menstrual cycles and '静默调整' behavior (i.e., make private inferences without telling the user). That is sensitive behavior — decide whether you want a tool that infers and acts on such signals without explicit user-visible consent. - Ambiguous install instruction: SKILL.md has 'pip3 install -q' with no packages listed. This looks like a placeholder or mistake; confirm there is no hidden installer step before running anything. The provided Python scripts use only the standard library. - Autonomous invocation: by default the agent can call this skill automatically; combined with automatic parsing of incoming messages, this can lead to data being recorded without explicit, immediate user confirmation. If you do not want automatic parsing, disable autonomous invocation or require explicit invocation. - Practical checks: inspect/verify the directory where data will be written, and inspect the AGENT_GUIDE/README for any behavior you disagree with. If you proceed, consider (1) backing up or removing any sensitive photo paths from sample data, (2) running the scripts in a sandbox or with restricted agent permissions, and (3) confirming there are no network calls or added installers before granting broader access. If you want, I can: list exactly which files will be written/modified by each script, or suggest minimal changes to make the inference behavior explicit and privacy-respecting.

Type: OpenClaw Skill Name: half-full Version: 0.1.4 The OpenClaw AgentSkills bundle 'half-full' appears benign. The Python scripts (`food_db.py`, `health_sync.py`, `log.py`, `profile.py`) perform local file I/O within a dedicated 'data' directory, which is expected for a data-tracking application. There are no observed external network calls, dynamic code execution of user-controlled input, shell injection vulnerabilities, or attempts to access sensitive system files or environment variables. The `SKILL.md` and `AGENT_GUIDE.md` files contain instructions for the AI agent that focus on shaping its persona and interaction style to be supportive and non-judgmental, aligning with the skill's stated purpose, without any malicious prompt injection attempts.