The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes and repeats untrusted user input to enhance model attention.
Ingestion points: The prompt string is ingested through the transform method and apply_prompt_repetition function in prompt_repetition_transformer.py.
Boundary markers: While the skill includes an applied_marker to prevent recursion, it does not wrap individual repetitions in delimiters to prevent the model from misinterpreting repeated malicious instructions as system-level directives.
Capability inventory: The included Python scripts are limited to string concatenation and regular expression matching; no high-risk capabilities such as subprocess calls, file writing, or network operations were detected.
Sanitization: The skill does not perform any validation, filtering, or escaping of the input prompt content before repeating it.

prompt-repetition