The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts located in the scripts/ directory, such as seedance_lint.py, higgsfield_memory.py, and build_index.py. The primary use case is a "Pre-flight Linter" (python3 scripts/seedance_lint.py --preflight --model seedance_2_0 "<prompt>") which validates prompt text before generation. This invocation interpolates user-supplied data into a shell command, representing a potential command injection surface if inputs are not properly escaped by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and reformulate untrusted user-supplied scene descriptions and generation failure feedback into structured "Seedance" prompts.
Ingestion points: User scene descriptions and error reports are ingested through the SKILL.md instructions and diagnostic workflows.
Boundary markers: The skill employs a "six-slot formula" and a "block scaffold" to provide structural delimiters for the generated prompts, though it lacks explicit XML-style sanitization tags for the raw user input passed to the linter command.
Capability inventory: The skill has the capability to execute shell commands (Python scripts) and perform file-write operations to local databases (e.g., filter-memory.json) to track prompt quality and filter performance.
Sanitization: The skill incorporates a robust "Pre-flight Linter" and a "Rewrite Playbook" specifically aimed at sanitizing and hardening prompts against external safety filter rejections by replacing risky keywords with professional cinematography descriptions.

higgsfield-seedance