The Agent Skills Directory

[DATA_EXFILTRATION]: The CLI tool in scripts/cdn.ts reads sensitive API credentials directly from the user's home directory at ~/.agents/.env.local. Accessing specific configuration files in hidden home directory paths is a sensitive file operation.
[COMMAND_EXECUTION]: The success criteria in SKILL.md instructs the agent to verify uploads by executing a curl command targeting a URL returned by the CDN service. This pattern is vulnerable to command injection if the remote service returns a malicious payload formatted to escape the shell command context.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data, including filenames and folder metadata, from an external CDN service without implementing boundary markers or sanitization.
Ingestion points: Data retrieved from the CDN API via the list, get, and upload commands in scripts/cdn.ts (e.g., asset filenames and folder paths).
Boundary markers: Absent. The instructions do not provide delimiters or instructions to ignore potential commands embedded in asset metadata.
Capability inventory: The skill possesses file system access (read/write) and network capabilities via fetch and curl.
Sanitization: There is no evidence of metadata sanitization or validation before the content is presented to the agent.

media-cdn