The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external 'source material' and 'explainer requests' to generate scripts and videos. This creates a surface where instructions embedded in source documents could attempt to influence agent behavior.
Ingestion points: Untrusted data enters via 'explainer-request' and 'source-material' as defined in SKILL.md.
Boundary markers: The skill does not explicitly define delimiters for external data, though it instructs the agent to exclude secrets and hidden instructions from its internal production records.
Capability inventory: The skill performs local file writes and executes workspace commands such as build and lint checks as described in references/production-and-review.md.
Sanitization: Instructions emphasize verifying claims against primary sources (code/docs) but do not detail programmatic sanitization of input text.
[COMMAND_EXECUTION]: The skill is designed to interact with a user's local development workspace, involving the execution of 'scaffold commands', 'lint', 'type', and 'production-build checks'. These operations are expected for the skill's purpose but involve executing commands based on project configuration.
[DATA_EXPOSURE]: The instructions include robust safety measures against credential exposure, specifically forbidding the recording, printing, or copying of credentials used for narration services and requiring explicit authority for any paid or external actions.

kata-explainer