The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using platform-provided binaries ($OD_BIN, $OD_NODE_BIN) to generate image assets. This involves interpolating variables like project IDs and model names into a command-line interface.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection and command injection because it ingests project metadata and user-provided briefs to construct shell command arguments without explicit sanitization instructions.
Ingestion points: Project metadata fields (imageModel, imageAspect, imageStyle) and the user's prompt brief are used to compose the final command in SKILL.md.
Boundary markers: No delimiters or 'ignore embedded instructions' warnings are present to isolate user-controlled data from the command structure.
Capability inventory: The skill uses shell execution capabilities via the $OD_BIN media tools (documented in SKILL.md) and is declared to have file-writing capabilities (fs:write) in open-design.json.
Sanitization: The instructions do not specify any validation, escaping, or filtering for the data interpolated into the --prompt or --model arguments, which could allow malicious input to break out of the shell command context.

image-poster