lytics-integrations
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, presenting a potential surface for indirect prompt injection. * Ingestion points: API responses in advise.md, connections.md, export-debug.md, jobs.md, setup.md, and webhook-templates.md; documentation content via WebFetch in webhook-templates.md. * Boundary markers: The confirmation-gate.md pattern requires manual user approval for all write operations, providing a defense against unintended actions. * Capability inventory: curl for API interaction, WebSearch and WebFetch for documentation retrieval. * Sanitization: Guidelines in api.md require URL-encoding of all user-supplied variables.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill identifies and accesses the configuration file ~/.lytics/accounts.toml to manage account-specific credentials. * Evidence: auth.md defines this path for storing API tokens. * Context: Access is limited to account resolution for synchronization tasks; logging of full tokens is strictly prohibited.
- [COMMAND_EXECUTION]: Executes shell commands to format data for API requests. * Evidence: webhook-templates.md uses python3 to URL-encode JSON payloads.
Audit Metadata