azure-ai-translation-document-py
Audited by Gen Agent Trust Hub on Feb 13, 2026
The skill azure-ai-translation-document-py and its associated acceptance-criteria.md file have been thoroughly analyzed for security vulnerabilities.
1. Prompt Injection: No patterns indicative of prompt injection (e.g., 'IMPORTANT: Ignore', 'Override your constraints') were found in either the skill description or the code examples.
2. Data Exfiltration: The skill involves handling sensitive information such as AZURE_DOCUMENT_TRANSLATION_KEY and Azure Blob Storage SAS tokens (AZURE_SOURCE_CONTAINER_URL, AZURE_TARGET_CONTAINER_URL). However, the skill explicitly recommends retrieving these from environment variables (os.environ) and emphasizes using SAS tokens with minimal required permissions. All network operations are directed towards official Azure domains (cognitiveservices.azure.com, blob.core.windows.net), which are trusted. File system access (with open(...)) is limited to reading and writing local documents for the skill's explicit translation purpose, not for accessing or exfiltrating sensitive system files.
3. Obfuscation: No obfuscation techniques (Base64, zero-width characters, homoglyphs, URL/hex/HTML encoding) were detected in the provided files.
4. Unverifiable Dependencies: The skill requires pip install azure-ai-translation-document. This is an official Azure SDK, which is considered a trusted external source. This dependency is noted as an informational finding but does not elevate the overall risk due to its trusted nature.
5. Privilege Escalation: No commands or instructions that attempt to escalate privileges (e.g., sudo, chmod 777, service installations) were found.
6. Persistence Mechanisms: No attempts to establish persistence (e.g., modifying shell configuration files, creating cron jobs) were detected.
7. Metadata Poisoning: The skill's name and description fields are clean and do not contain any hidden malicious instructions.
8. Indirect Prompt Injection: While any skill processing external content (like documents) could theoretically be susceptible to indirect prompt injection if that content were passed to an LLM, this skill is for document translation and does not appear to involve LLM processing of the document content itself. The skill's design does not introduce this vulnerability directly.
9. Time-Delayed / Conditional Attacks: No conditional logic or time-based triggers for malicious behavior were identified.
Adversarial Reasoning: The references/acceptance-criteria.md file actively promotes secure coding practices, such as avoiding hardcoded credentials and ensuring proper SAS token usage. This indicates a strong security-conscious design. The skill's functionality aligns with its stated purpose, and there are no suspicious discrepancies. The use of official Azure SDKs further reduces risk.
Conclusion: The skill is well-designed with security best practices in mind, particularly regarding credential handling and interaction with trusted cloud services. The only external dependency is from a trusted source. Therefore, the skill is deemed SAFE.