Security audit
Security checks for vulnerabilities and agentic risk
This simple skill writes two local files and fetches one website; its side effects are clear, but users should avoid running it where a.txt or b.txt contain important data.
Install only if you want a demo-style skill that creates or replaces a.txt and b.txt in the active workspace and contacts www.bytedance.com. Check for existing files with those names before running it in a real project directory.
The skill explicitly instructs the agent to overwrite a.txt and b.txt if they already exist, but provides no warning, confirmation step, or safeguard against destroying existing workspace data. In an automated agent context, this can cause unintended data loss, especially because the current workspace may contain user-created files with those names.
The instruction to proceed in the current workspace and not ask the user for a target path encourages autonomous file-writing decisions without validating the destination. This increases the risk of unintended modifications in whatever directory happens to be active, especially when combined with unconditional overwrite behavior.
- `b.txt` should contain a concise summary of the fetched page content.
- If either file already exists, overwrite it.
- Keep the filesystem changes simple and local. Network access is only needed for the `web_fetch` step.
- Do not ask the user for a target path unless they explicitly want behavior different from the current workspace directory.
- If you need to confirm the result, inspect the directory after writing with `ls` or equivalent.
- If `web_fetch` fails, report that clearly instead of inventing a summary.
The skill manifest is broadly phrased and does not clearly constrain when it should be invoked beyond a general user desire. Because it combines local file writes and a network fetch, ambiguous triggering increases the chance an agent will invoke it in situations the user did not specifically intend, causing unintended side effects and external requests.
The manifest explicitly instructs the agent to write two files in the workspace and perform a network fetch, but it does not warn the user about these side effects. This can lead to silent modification of local state and outbound network access, which is risky in environments where file persistence, provenance, or network usage must be tightly controlled.
No suspicious patterns detected.