The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands constructed from user-supplied arguments or inferred from project configuration files. This allows for arbitrary command execution on the host system.
[DATA_EXFILTRATION]: The skill automates the creation of a public ingress tunnel to the local machine. This allows anyone with the generated URL to access local services and potentially exfiltrate sensitive data or environment information.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by trusting the contents of project files or repository documentation to define its execution plan.
Ingestion points: Reads package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod, and repository documentation.
Boundary markers: None. The skill does not use delimiters to separate trusted instructions from untrusted file content.
Capability inventory: Full Bash access, allowing for arbitrary process creation and file system modification.
Sanitization: None. Data from files is used directly to determine the dev/start command.
[EXTERNAL_DOWNLOADS]: The workflow involves downloading and installing third-party binaries (cloudflared, ngrok). While these are well-known services, the automated installation process via shell commands poses a risk if the installation source or integrity is not verified.

live-preview