The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and orchestrates the execution of user-defined test prompts and trigger queries, which represents a potential surface for indirect prompt injection if malicious test cases are supplied.
Ingestion points: Test prompts in evals/suites/*.json, trigger queries in trigger-queries.json, and direct user inputs during the skill interview.
Boundary markers: No explicit instruction-level delimiters (e.g., [DATA]...[/DATA]) are specified for the sub-agent runs used in the evaluation process.
Capability inventory: The skill uses npm scripts for local execution, write_to_file for saving outputs, and sub-agent spawning for benchmarking.
Sanitization: No content sanitization or validation of the user-provided prompts is mentioned in the instructions.
Contextual Note: This surface is inherent to the skill's purpose as an evaluation framework for AI instructions and is not considered a malicious finding in this context.

skill-creator