The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/casting_agent.py uses subprocess.run to execute internal tasks and sanity checks using the current python executable and relative paths.\n- [PROMPT_INJECTION]: The skill ingests untrusted story context and entity descriptions (Category 8 surface). Ingestion points: story_contract.md and visual_entities.json. Boundary markers: None. Capability inventory: Subprocess execution, file writes, and delegated network search.\n- [DATA_EXFILTRATION]: The _write_package_inputs function reads local files based on paths provided in input JSON documents, which could lead to unauthorized data exposure if input files are malicious.

casting-agent