The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted customer data from service tickets and survey responses. In references/nps-methodology.md, a prompt template is provided for LLM analysis of customer comments (Comment: {text}). This represents an attack surface for indirect prompt injection.\n
Ingestion points: Untrusted data enters via ticket descriptions and NPS open-text responses, described in SKILL.md and references/nps-methodology.md.\n
Boundary markers: The prompt template for analyzing detractor comments in references/nps-methodology.md lacks explicit delimiters or instructions to ignore embedded commands.\n
Capability inventory: The skill includes Python scripts for data processing and keyword extraction; no high-risk capabilities like file system modifications or external network operations are enabled for the agent based on this data.\n
Sanitization: A preprocess function in references/ticket-text-mining.md removes URLs, emails, and ticket IDs, providing a basic level of filtering for typical injection vectors.\n- [EXTERNAL_DOWNLOADS]: The references/ticket-text-mining.md file references the installation of several external Python packages, including bertopic, sentence-transformers, umap-learn, hdbscan, vaderSentiment, and ckip-transformers. It also suggests fetching a pre-trained model (paraphrase-multilingual-MiniLM-L12-v2) from Hugging Face for topic clustering.

cs-analytics