The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses tools like java_launch to execute local processes and instructs the agent to run commands such as jabswitch -enable and robomotion describe package. These are documented functionalities that require user visibility and interaction.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests text and structural data from external Java applications' UI trees, which constitutes a potential injection surface.\n
Ingestion points: UI element information is retrieved through tools like java_snapshot, java_find, and java_read_table mentioned in SKILL.md.\n
Boundary markers: The agent interacts with elements using generated reference identifiers (e.g., @e1), which helps separate UI content from the automation structure.\n
Capability inventory: The skill provides capabilities to launch processes, simulate keyboard and mouse input, and generate automation code, as described in SKILL.md.\n
Sanitization: While no specific text filtering is mentioned, the skill implements safety checks by requiring the agent to obtain user confirmation before performing sensitive actions or finalizing code.

exploring-java