Lobster Radio – Free Local AI Radio
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill’s radio/TTS purpose is coherent, but it deserves review because it downloads and loads external model code and can create recurring message-push tasks.
Review the install and model-loading steps before use. This skill appears aligned with generating local TTS radio, but you should approve any model downloads, prefer pinned/audited model revisions, avoid storing sensitive preferences, and verify any scheduled push destinations.
Findings (6)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
First use may pull large external model artifacts onto the user’s machine, and the exact reviewed code/data may differ over time if the upstream source changes.
The skill says it may automatically download the TTS model on first generation, but the artifacts do not show commit pins, hashes, or integrity checks for the HuggingFace/ModelScope model sources.
Skill会在首次生成电台时自动检测并下载模型
Pin model revisions, verify checksums or signatures, and ask the user before automatic downloads.
A changed or compromised model repository could run code locally when the model is loaded.
The guide instructs loading model code with `trust_remote_code=True`, which can execute custom Python code from the downloaded model repository.
AutoModelForCausalLM.from_pretrained(
model_name,
trust_remote_code=True,Avoid `trust_remote_code=True` where possible, or pin and audit the exact model revision before loading it.
If enabled, the skill can keep producing and sending content on a schedule until the task is changed or removed.
The skill documents creating recurring scheduled tasks that can automatically generate and announce radio content.
openclaw cron add \ --name "每日科技电台" \ --cron "0 8 * * *" \ --session isolated \ --message "生成科技新闻电台" \ --announce \ --channel feishu
Only create schedules intentionally, review destination channels, and use the platform’s cron list/delete controls to manage them.
Saved preferences may persist across sessions and influence later outputs.
The skill stores preferences in persistent memory so they can affect future radio generation.
用户偏好（订阅标签、常用音色等）： - **OpenClaw**: 存储在MEMORY.md中 - **LobsterAI**: 存储在持久化记忆中
Do not store sensitive personal information as radio preferences, and review or clear persistent memory if needed.
Users might assume no network or cloud-dependent processing is involved, even though news search, model downloads, or the platform LLM may use external services.
The privacy/offline wording is broader than other artifacts showing web-search use, platform LLM content generation, and model downloads.
🔒 **数据隐私**: 所有处理在本地完成 ... ✅ **可离线使用**: 无网络依赖
Clarify that local processing mainly applies to TTS, and disclose when search, LLM providers, or model downloads use the network.
The documented command may fail or rely on code not included in the reviewed artifact set.
The documentation repeatedly references `scripts/generate_radio.py`, but the provided file manifest does not list that script, creating a review/provenance gap for the main generation entry point.
python scripts/generate_radio.py --topics "人工智能" --tags "科技"
Include the referenced generation script in the package or update the documentation to the actual entry point.