The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script (pinterest.py) using the Bash tool for all API interactions. It also includes a find command in the SKILL.md instructions to dynamically locate the script file within the /tmp directory structure, which could lead to path hijacking if an attacker places a malicious script in a matching directory pattern.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes untrusted user input and interpolates it into shell commands executed via the Bash tool. Ingestion points: User-provided content such as Pin titles, descriptions, and board IDs are ingested through the agent's interaction context in SKILL.md. Boundary markers: None. The instructions do not provide explicit delimiters or warnings to the agent about handling embedded instructions in user data. Capability inventory: The skill has access to the Bash tool for local command execution and the publish_artifact tool. The bundled script can make network requests to api.pinterest.com using the user's PINTEREST_TOKEN. Sanitization: The Python script uses argparse to safely handle arguments internally and validates that image URLs use HTTPS. However, there is no explicit instruction for the agent to sanitize or escape strings when constructing the initial shell command.

pinterest