customerio-observability
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements logic to ingest and process data from external Customer.io API operations and webhooks for the purpose of logging and metrics. This creates a surface where malicious payloads embedded in event data could influence downstream processing.\n
- Ingestion points: Data enters the context via
InstrumentedCioClientmethods (SKILL.md) and therecordWebhookMetricsfunction (SKILL.md), which handle external API responses and webhook payloads.\n - Boundary markers: No specific boundary markers for LLM context separation are present, though the skill provides guidance on ignoring specific sensitive fields in its examples.\n
- Capability inventory: The skill is designed to operate with tools including
Write,Edit, andBash.\n - Sanitization: The skill includes strong mitigation instructions, providing explicit PII redaction logic using
pino(SKILL.md) and a dedicatedsanitizeForLoggingfunction (references/implementation-guide.md) targeting fields such as email, phone, and password.\n- [COMMAND_EXECUTION]: The skill provides troubleshooting commands that execute shell operations for local environment verification.\n - Evidence: The troubleshooting section (references/implementation.md) includes
curl -s localhost:9090/metrics | grep customerio.\n - Context: These commands are restricted to
localhostand are standard practices for verifying observability stacks during development.
Audit Metadata