The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an evaluation framework where an LLM judge processes potentially untrusted data from workflow outputs and external datasets. This represents a vulnerability surface where malicious content within the evaluated data could attempt to influence the judge's scoring behavior.\n
Ingestion points: Data enters the evaluation context from Data Tables (referenced by dataTableId) and workflow node results (captured by actualAnswer expressions) as described in SKILL.md and references/config-eval-playbook.md.\n
Boundary markers: The instructions utilize n8n expressions to select specific data fields but do not define mandatory prompt isolation techniques or boundary markers to prevent the judge model from following instructions embedded within that data.\n
Capability inventory: The skill leverages the eval-config tool for creating and updating evaluation settings and the data-tables tool for dataset management.\n
Sanitization: There is no requirement in the skill instructions to sanitize, filter, or escape the content being evaluated before it is presented to the LLM judge model.

config-evals