connecting-to-data-source
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill constructs and executes various AWS CLI commands using variables derived from user input, such as connection names, hostnames, and project identifiers (e.g.,
aws glue create-connection,aws glue test-connection). The instructions lack sanitization or validation logic for these inputs, which could allow for command injection if a user provides strings containing shell metacharacters. - [REMOTE_CODE_EXECUTION]: In
references/troubleshooting.md, the skill instructs the agent to generate a Python script (test-connection.py) and run it as an AWS Glue ETL job usingaws glue create-jobandaws glue start-job-run. This represents dynamic code generation and execution within the user's AWS infrastructure. - [PROMPT_INJECTION]: The skill ingests untrusted data from the user during the "Gather Connection Hints" phase (Step 3) and interpolates this data into commands and prompts. There are no boundary markers or instructions to treat this content as untrusted, creating an attack surface for indirect prompt injection where malicious inputs could influence the agent's behavior.
- [DATA_EXFILTRATION]: The workflow requires the agent to handle sensitive credentials, such as database passwords and GCP service account keys, to facilitate their storage in AWS Secrets Manager. While the skill advises using Secrets Manager and warns against logging, the agent still performs operations like base64 encoding these secrets on the command line, which may expose them in process trees or command history.
Audit Metadata