AgenticWorkflowGuardrails

Optional timeout after which the Agent0 automation will be aborted. Defaults to ten minutes.

Optionally restrict which tools are allowed. By default, all tools are allowed to be used. You can use the * character to allow all tools matching a pattern.

Optionally enforce that the Agent0 automation has executed these tool calls. If the agent is concluding is automation, but hasn't called these tools, the agent receives a follow-up prompt to call these tools. This auto-correction of the LLM behavior is only attempted once.

This is useful if you wanna ensure that the agent makes specific tool calls to external systems, e.g., to leave a message on Slack or GitHub.

How long to wait before firing if the same trigger fires repeatedly. This is useful to avoid redundant automation executions when triggers fire in rapid succession. Defaults to 10s, which is also the minimum: a shorter non-zero value is raised to 10s. Set to "0s" to disable debouncing.

Pre-configure tool parameters that are enforced at invocation time. When a listed tool is called by the agent, these parameters are merged into the tool's arguments, overriding any values the LLM may have provided. Keys can be either fully qualified with namespace (e.g., "dash0.slackSendMessage") or bare tool names without namespace (e.g., "slackSendMessage"). Namespaced keys take precedence and allow disambiguation when multiple integrations expose tools with the same name. Values are objects of key-value pairs merged into tool arguments. Values support {{variable}} substitution using built-in variables, constants, and trigger variables.

Optional rolling-window credit budget across all runs of this workflow. Checked before each run starts: a new run is not started when the budget is already exhausted. A run already in flight is not aborted when the budget is crossed, so spend can exceed the budget. Leave unset for no budget.

Per-automation overrides for MCP server tools. Each entry sets whether a specific MCP tool is available to this automation, overriding the tool's enabled/disabled state in the organization's MCP integration settings. Tools not listed here fall back to their integration-settings default. This lets an automation enable a tool that is disabled in settings, or disable a tool that is enabled there, scoped to this automation only.

The custom (organization-authored) skills this automation may use, by skill name (lowercase alphanumerics separated by single hyphens, e.g. cost-report). A run can only list and load the custom skills named here, and still only those the automation's creator may read. When absent or empty, the automation uses no custom skills. Dash0's built-in skills are not affected. Naming a skill as /name in the prompt does not add it to this list.