The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data (shot intents and storyboards) which are then used to construct prompts for a video generation CLI.
Ingestion points: The workflow in SKILL.md requires taking a 'one-line shot intent from the storyboard' as a primary input.
Boundary markers: The instructions provide a 'Fill-in template' to structure the output, but there are no explicit instructions to the agent to sanitize or ignore potentially malicious instructions embedded within the intent strings.
Capability inventory: The skill performs shell command execution through the higgsfield CLI tool, which is a high-privilege capability in the context of user-controlled inputs.
Sanitization: There is no evidence of input validation or escaping for the shot intent before it is interpolated into the CLI prompt string.
[COMMAND_EXECUTION]: The skill relies on and provides detailed documentation for executing shell commands to interact with the Higgsfield API.
Evidence: The 'Running the gen on Higgsfield' section details specific command patterns such as higgsfield generate create seedance_2_0 --prompt "...".
Context: These commands are documented as the standard procedure for the skill's functionality and involve local tool execution and network communication with the service provider.

higgsfield-video-prompt