The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text input via the script_text payload. It mitigates potential instruction injection by establishing strict 'Boundaries' that forbid the agent from rewriting dialogue, inventing events, or inferring visual details not in the source.
Ingestion points: payload.script_text defined in SKILL.md and validated in schemas/input.schema.json.
Boundary markers: Present; the Boundaries section in SKILL.md provides explicit negative constraints.
Capability inventory: No capabilities detected. The skill uses no tools, shell commands, or network operations.
Sanitization: Input is validated against a JSON schema for length, but no content filtering is applied.
[EXTERNAL_DOWNLOADS]: The schema files reference the official json-schema.org domain for standard schema definitions. This is a well-known service used for metadata validation and does not involve remote code execution.

structure-screenplay