The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill follows instructions from the open-design MCP server to author files.
Ingestion points: The output of the prepare_open_design_brief tool in SKILL.md is used as the primary source for file-editing instructions.
Boundary markers: There are no delimiters or 'ignore previous instructions' warnings applied to the data received from the tool output.
Capability inventory: The agent is instructed to use its own file-editing tools to create content based on external instructions, involving file write operations.
Sanitization: The skill lacks logic to validate or escape the instructions returned by the MCP tool before execution.
[PROMPT_INJECTION]: User input via $ARGUMENTS is directly embedded into the tool's input brief.
Evidence: The brief construction Brief: "$ARGUMENTS" in SKILL.md allows unvalidated user text to be passed to the prepare_open_design_brief tool, creating a surface where an attacker could potentially manipulate the tool's parameters or the agent's behavior through the brief content.

od-prototype-social-carousel