← 返回 Skills 市场
zhuyinzhuyin

STH Video Template Generation

作者 Yin Zhu · GitHub ↗ · v2.0.0 · MIT-0
cross-platform ⚠ suspicious
214
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install sth-video-gen
功能描述
Generate vertical 9:16 videos for Sing The Hook song templates using a two-stage pipeline with MCP, trimming, uploading to GCS, and database updates.
安全使用建议
Key points to verify before installing or running: - Mismatched declarations: The registry says "no env vars" but SKILL.md asks for many sensitive variables (DB creds, MCP API key, GCS key path). Ask the author which is authoritative. Do not hand over secrets until you confirm the code will use them correctly. - Code vs env: Inspect/modify the code to ensure DB/MCP/GCS configuration is read from secure environment variables (or a config file) rather than using hardcoded defaults (the current DB_CONFIG and MCP_CONFIG are hardcoded). If you expect it to use STH_DB_* / STH_MCP_* variables, confirm the code reads os.environ for those keys. - Run in an isolated sandbox: Because the scripts will execute psql, curl, ffmpeg/ffprobe and update a PostgreSQL table, test the skill in a throwaway environment (or with a read-only test DB and fake MCP credentials) before providing production credentials. - Verify external endpoints and credentials: The code targets an external MCP endpoint (https://kansas3.8glabs.com/mcp). Confirm that endpoint is trusted and that the MCP API key will be scoped/rotated appropriately. The skill also uploads to GCS — prefer a service account with minimal permissions (only the target bucket) and avoid broad-scoped keys. - File paths and data exposure: The scripts reference /root/.openclaw/media/inbound and write logs and state under the skill directory. Ensure those paths are acceptable and contain no sensitive data you don't want processed. Consider running with a limited filesystem view. - Review and harden: If you plan to use this, update the code to read credentials from environment variables, validate/escape template IDs before building SQL, and limit polling durations/worker counts. Remove or document any behavior that writes notifications or uses Telegram. If you cannot confirm these items with the author, treat the skill as untrusted and avoid supplying real DB or cloud credentials.
功能分析
Type: OpenClaw Skill Name: sth-video-gen Version: 2.0.0 The skill bundle implements a video generation pipeline but contains multiple critical SQL injection vulnerabilities across several files, including sth_video_generator.py and batch_processor.py, where template IDs from user-provided CSV inputs are directly interpolated into database queries. The scripts also perform high-risk operations such as executing shell commands (psql, ffmpeg, curl) and uploading data to Google Cloud Storage. While these actions appear aligned with the stated purpose of the 'Sing The Hook' service, the lack of input sanitization and the broad use of system commands present a significant security risk without clear evidence of intentional malice.
能力评估
Purpose & Capability
The skill claims to generate STH videos via MCP and upload to GCS (consistent with included scripts). However the registry metadata lists no required env vars/binaries while SKILL.md requires many sensitive values (DB host/credentials, MCP endpoint & API key, GCS key path). Worse: the Python code largely uses hardcoded DB/MCP configuration constants (localhost/dev_mobile/openclaw and MCP endpoint) rather than reading the SKILL.md-listed STH_* environment variables — a direct mismatch that is not explained and may cause the skill to act on local resources unexpectedly.
Instruction Scope
SKILL.md gives a narrow runtime flow (ask for CSV, run sth_video_generator.py). The actual code goes beyond that: it runs psql, curl, ffprobe/ffmpeg, downloads audio to temp files, writes logs and state files in the skill and /root/.openclaw/media paths, polls external MCP endpoints, and updates a database table. The SKILL.md does not document some hardcoded file paths (e.g., /root/.openclaw/media/inbound) or the precise behavior of the background polling and file writes. The instructions also ask for env vars that the code doesn't appear to read, meaning the runtime behavior may rely on hardcoded defaults instead of provided secrets.
Install Mechanism
No install spec is provided (instruction-only), which lowers supply-chain risk. The SKILL.md recommends installing google-cloud-storage and requires ffmpeg/psql/curl to be present on host. Because the code is included in the bundle (not downloaded at runtime) there's no remote arbitrary download, but the skill depends on system binaries and a Python package; the skill will call external binaries (curl/psql/ffprobe/ffmpeg) and the google-cloud library, so you must ensure those are installed in a controlled environment.
Credentials
SKILL.md requests many sensitive environment variables (DB host/port/name/user/password, MCP endpoint/api key, GCS service-account path). That is proportionate for the declared pipeline — except the code does not actually read those STH_* env vars and instead uses hardcoded DB_CONFIG/MCP_CONFIG values. This mismatch is concerning: either the SKILL.md is out of date or the code will ignore user-provided credentials and act on local defaults (localhost DB, empty DB password, hardcoded MCP endpoint). Requesting a GCS key file path and an MCP API key is expected for uploads, but you must confirm the code will use the credentials you provide rather than ignoring them.
Persistence & Privilege
The skill is not always-enabled and does not ask for platform-level privileges. It writes state and log files inside its skill directory (raw_video_urls.json, logs, pending_notifications.txt) and accesses files under /root/.openclaw/media/inbound in several scripts. It also spawns long-polling loops (poll_video_job) that can run for many minutes. Allowing autonomous invocation plus DB write capability increases blast radius; while autonomous invocation is normal, combined with database updates and external network calls this merits careful credential and environment controls.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install sth-video-gen
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /sth-video-gen 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v2.0.0
Clean v2 release
元数据
Slug sth-video-gen
版本 2.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

STH Video Template Generation 是什么?

Generate vertical 9:16 videos for Sing The Hook song templates using a two-stage pipeline with MCP, trimming, uploading to GCS, and database updates. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 214 次。

如何安装 STH Video Template Generation?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install sth-video-gen」即可一键安装,无需额外配置。

STH Video Template Generation 是免费的吗?

是的,STH Video Template Generation 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

STH Video Template Generation 支持哪些平台?

STH Video Template Generation 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 STH Video Template Generation?

由 Yin Zhu(@zhuyinzhuyin)开发并维护,当前版本 v2.0.0。

💬 留言讨论