topic-analysis
Pass
Audited by Gen Agent Trust Hub on Jul 4, 2026
Risk Level: SAFENO_CODEEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill defines a legitimate data science workflow for identifying latent topics in text documents. No evidence of obfuscation, malicious instructions, or data exfiltration was found.
- [NO_CODE]: The skill consists purely of instructional documentation (SKILL.md) and does not include any executable scripts or binary files.
- [EXTERNAL_DOWNLOADS]: The procedure mentions well-known machine learning libraries (BERTopic, gensim, sentence-transformers) and official service providers (OpenAI, Voyage). These are established technologies used for the skill's primary purpose.
- [PROMPT_INJECTION]: The skill processes external data (corpora), which inherently introduces a surface for indirect prompt injection. However, the instructions incorporate data cleaning steps that mitigate risk.
- Ingestion points: JSONL, CSV, and folders of text files (SKILL.md)
- Boundary markers: None specified in the instructions.
- Capability inventory: LLM-assisted labeling of topic clusters.
- Sanitization: The procedure includes cleaning steps such as lowercasing, stripping URLs, and removing stopwords.
Audit Metadata