The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests image and mask URLs from external sources, which represent an untrusted data surface where embedded instructions could potentially influence the model's output.
Ingestion points: image and mask_image parameters passed to the runcomfy CLI as described in SKILL.md.
Boundary markers: The documentation explicitly instructs agents to only ingest URLs provided by the user for the specific task.
Capability inventory: The skill uses a scoped Bash tool to execute the runcomfy CLI for image processing.
Sanitization: Prompt content is passed as a JSON string to the CLI, which avoids shell expansion and mitigates shell injection risks.
[COMMAND_EXECUTION]: The skill performs command execution via the runcomfy CLI. This is controlled by the allowed-tools configuration which restricts the agent's capability to only the necessary vendor-provided command.

image-inpainting