airbyte-agent
Fail
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides an installation command for the
airbyte-agentCLI usingcurl -fsSL https://airbyte.ai/install.sh | bash. This executes a script from the vendor's official domain directly in the shell. While this is a standard installation pattern for the service, it involves the execution of remote code.\n- [COMMAND_EXECUTION]: The skill relies on executing theairbyte-agentCLI to perform management tasks. It also modifies the local configuration file at~/.airbyte-agent/settings.jsonwhen setting a default workspace via theworkspaces usecommand.\n- [DATA_EXFILTRATION]: The skill's core functionality is to read data from external SaaS products (e.g., Salesforce, HubSpot) through Airbyte connectors and ingest it into the agent's context, which constitutes the movement of external data into the local environment.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes data fetched from external, potentially untrusted SaaS sources.\n - Ingestion points: Data returned from
airbyte-agent connectors executeoperations.\n - Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded prompts within the external content.\n
- Capability inventory: The skill can perform write actions, delete connectors, and modify local configuration files.\n
- Sanitization: There is no mention of validating or escaping retrieved data before it is processed by the agent.
Recommendations
- HIGH: Downloads and executes remote code from: https://airbyte.ai/install.sh - DO NOT USE without thorough review
Audit Metadata