The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references numerous official documentation sites and repositories for observability tools. It provides example commands for installing legitimate CNCF projects and vendor SDKs (e.g., helm install beyla, sentry-cli upload-sourcemaps). All sources are trusted organizations or well-known services.
[COMMAND_EXECUTION]: The documentation includes many examples of CLI usage for DevOps tasks (e.g., kubectl apply, chronyc tracking, promtool check rules). These are provided as educational snippets and configuration guides for the user to manage their infrastructure and are not executed automatically by the agent for malicious purposes.
[DATA_EXFILTRATION]: The skill provides guidance on protecting sensitive data, including PII redaction rules for logs and traces (GDPR/PIPA compliance). It recommends masking IP addresses and using salted HMACs for user identifiers. No evidence of unauthorized data transmission or credential harvesting was found.
[PROMPT_INJECTION]: The instructions and keyword rules are well-defined for intent classification (setup, migrate, investigate, etc.). There are no attempts to override safety filters or bypass initial instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest telemetry data such as logs and traces for forensics. While this represents a surface for indirect injection, the skill's capabilities are limited to analytical reasoning (SELECT, INFER) and generating documentation or configuration templates, minimizing the risk of exploitation.

oma-observability