amazon-bedrock
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates building RAG (Retrieval-Augmented Generation) applications that ingest data from external sources (S3, SharePoint, etc.). The instructions include detailed guidance on using Bedrock Guardrails and 'guardContent' blocks to isolate and filter untrusted inputs, mitigating potential injection risks.\n
- Ingestion points: Data retrieved from Knowledge Bases via the 'Retrieve' and 'RetrieveAndGenerate' APIs.\n
- Boundary markers: Use of 'guardContent' blocks is explicitly recommended for wrapping untrusted content.\n
- Capability inventory: Capabilities include AWS CLI commands, boto3 scripting, and tool invocation via Lambda shims.\n
- Sanitization: Recommends pre-ingestion PII redaction and post-generation filtering using Guardrails.\n- [DATA_EXFILTRATION]: The provided 'fetch_bedrock_agent.py' script performs local discovery by aggregating metadata about the AWS environment, including IAM policy documents, account IDs, and resource configurations, into a 'source-agent.json' file for migration. The script adopts security best practices by restricting file permissions ('chmod 0600') and alerting the user to the sensitivity of the data.\n- [DYNAMIC_EXECUTION]: The skill uses local templates to generate Python deployment scripts and Lambda shims. These are used to bridge API differences during migration or to set up managed payment services. The logic follows standard software development patterns and avoids unsafe execution of untrusted code.\n- [EXTERNAL_DOWNLOADS]: The skill references the '@aws/agentcore' package from the official npm registry and the 'boto3' library from PyPI, both of which are standard tools for interacting with AWS services.
Audit Metadata