The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute system commands, including searching the file system and running Python scripts via the python3 -c interface.
[REMOTE_CODE_EXECUTION]: The instructions explicitly direct the AI agent to perform string substitution of user-provided arguments (INPUT_PATH, OUTPUT_PATH, and TEXT_PROMPT) into Python code templates. Because the input is not sanitized or escaped, a crafted input containing Python delimiters (e.g., quotes, semicolons, or parentheses) can break out of the intended string context to execute arbitrary Python code and system commands.
[EXTERNAL_DOWNLOADS]: The skill is configured to download pre-trained AI models from Hugging Face (a well-known repository for machine learning assets) during its initial execution.
[PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection via user-supplied text prompts and file paths.
Ingestion points: User input enters the system through the $1 (raster path) and --text (GroundedSAM prompt) arguments.
Boundary markers: No boundary markers or 'ignore' instructions are provided to the agent to protect the execution context from malicious content in the user input.
Capability inventory: The skill has access to the Bash tool, allowing it to execute arbitrary commands, read/write files, and perform network operations.
Sanitization: There is no evidence of input validation or sanitization; the instructions encourage direct interpolation of untrusted strings into executable Python commands.

detect-objects