The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local scripts using uv run (e.g., scripts/scaffold.py, scripts/storyboard.py) and standard shell commands like touch, export, and printf. These operations are used for environment setup, file scaffolding, and creating sentinel files to signal task completion within a multi-agent pipeline.
[PROMPT_INJECTION]: The skill uses directive language such as "iron rules," "red lines," and "non-negotiable" to enforce structural and stylistic constraints. It explicitly defines an external local file (references/shanyin/screenwriting-master/SKILL.md) as the primary authority for writing craft. This is an architectural design for modular agent skills rather than an attempt to bypass system safety guidelines.
[DATA_EXFILTRATION]: The skill interacts with local project files (e.g., lore.md, cast.json) and environment variables. There are no indications of network requests, hardcoded credentials, or attempts to access sensitive system directories like .ssh or .aws.
[REMOTE_CODE_EXECUTION]: All script executions and file references are restricted to the local project structure. No external packages are downloaded from untrusted sources, and no remote code is piped into the shell.

spark-video-screenwriter