The Agent Skills Directory

[Data Exposure & Exfiltration] (HIGH): The script scripts/generate-mock.sh accepts an unvalidated $ENDPOINT variable and passes it directly to curl. An attacker could exploit this to perform Local File Inclusion (LFI) by providing a URI such as file:///etc/passwd, causing the agent to read and save sensitive local files to the Mocks/ directory.
[Indirect Prompt Injection] (HIGH): The skill exposes a data ingestion surface that lacks sanitization or boundaries.
Ingestion points: scripts/generate-mock.sh (via curl fetch from remote/local URLs).
Boundary markers: Absent; the fetched content is written directly to Mocks/response.json.
Capability inventory: The skill possesses network read (curl) and file system write (redirection) capabilities.
Sanitization: None; there is no validation of the URL scheme or the content returned by the endpoint.
[Command Execution] (MEDIUM): The skill includes an executable shell script that performs network and file system operations. While intended for mock generation, its lack of input validation increases the risk of the agent being coerced into performing unintended system actions.

swift-networking