The Agent Skills Directory

[PROMPT_INJECTION]: The user input provided via $ARGUMENTS is directly placed into the prompt without any boundary markers or ignore instructions warnings. This allows for direct prompt injection where a user could attempt to override the agent's instructions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from external customer support tickets. 1. Ingestion points: User-provided text in $ARGUMENTS and ticket message threads retrieved from Gorgias via the GORGIAS_GET_TICKET command (SKILL.md). 2. Boundary markers: Absent. The data is treated as direct instructional context. 3. Capability inventory: The skill has the ability to execute shell commands via the composio CLI (SKILL.md). 4. Sanitization: No validation or filtering is performed on the input text before processing or when constructing analysis outputs.
[COMMAND_EXECUTION]: The skill utilizes the Bash shell to run composio commands. While this is the intended mechanism for data retrieval in this context, the lack of input sanitization creates a surface for potential misuse if the agent is misled by malicious input.

sentiment-check