The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting and processing untrusted data from external Skyrim plugin files.
Ingestion points: The skill uses housecarl_cross_plugin_query and housecarl_batch_record_detail in SKILL.md to fetch record fields such as 'Name', 'EditorID', and other string-based data from .esp and .esm files.
Boundary markers: There are no instructions to use delimiters or 'ignore embedded instructions' warnings when these field values are interpolated into the agent's context or passed to subagents in the 'Recipe: the fan-out' section.
Capability inventory: The skill includes tools like housecarl_bulk_apply and housecarl_forward_record (SKILL.md) which allow the agent to modify, forward, and rebuild records in plugin files based on the processed data.
Sanitization: The skill does not define any sanitization, escaping, or filtering logic for external record content before it is used to drive the extraction or patching logic.

bulk-record-jobs