The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external customer tickets (via the Gorgias API) and user-supplied arguments without employing boundary markers or sanitization. This could allow an attacker to embed instructions within a support ticket that attempt to manipulate the agent's translation or drafting logic.
Ingestion points: User input through $ARGUMENTS and external ticket content retrieved using the GORGIAS_GET_TICKET tool in SKILL.md.
Boundary markers: No delimiters (e.g., XML tags or triple quotes) are used to separate instructions from the untrusted ticket data.
Capability inventory: The skill utilizes composio search and composio execute commands within a Bash environment to interact with external services.
Sanitization: There is no evidence of input validation or escaping for the ticket content before it is processed by the language model.
[COMMAND_EXECUTION]: The skill workflow involves executing shell commands via the Composio CLI (composio search, composio execute) to fetch ticket details. While these are part of the intended integration with the Gorgias platform, the use of user-supplied arguments to populate command parameters (ticket_id) without explicit validation represents a potential surface for command manipulation if the agent does not strictly parse the input as an ID.

translate-ticket