The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an interface for processing external telemetry data (logs, traces, RUM events) that could potentially contain malicious or adversarial instructions.
Ingestion points: Untrusted data enters the agent context through tools like search-logs, search-traces, and search-rum (SKILL.md).
Boundary markers: The documentation does not specify mandatory delimiters or 'ignore' instructions for the content of processed logs.
Capability inventory: The skill possesses capabilities to execute SQL/SPL queries and mutate system state via create-alert and update-dashboard (SKILL.md).
Sanitization: The skill provides and recommends dedicated validation tools (validate-alert-json, validate-dashboard-json, validate-sql) to check inputs before execution or state modification (SKILL.md).
[COMMAND_EXECUTION]: The skill provides tools for executing arbitrary database and log-processing queries.
Evidence: Tools execute-sql and execute-spl allow raw query execution against ClickHouse or log pipelines (SKILL.md).
Context: These are documented as 'escape hatches' and are subject to server-side RBAC and scope restrictions (SKILL.md).
[DYNAMIC_EXECUTION]: The skill facilitates the dynamic assembly and resolution of queries at runtime using variables and templates.
Evidence: Dashboard variables and macros (e.g., $__timeFilter, $__clusters) are substituted into query strings before execution by tools like run-dashboard-panel (SKILL.md).

kubesense-mcp