Orchestrate and run Routines
A Routine combines Sprites into a reusable governance workflow. It controls node order, input and output mappings, failure handling, and execution records. The same Sprite can be reused by multiple Routines without changing its responsibility.
Plan a Routine
Map business inputs, steps, and acceptance results before creating nodes. Each node invokes one Sprite and performs a clearly stated action. Independent read-only checks can run in parallel; data and write dependencies require sequential execution.
| Element | Definition |
|---|---|
| Entry parameters | Objects, space, or filters supplied by the caller |
| Node | One Sprite invocation |
| Parameter mappings | Entry inputs or upstream outputs passed to a node |
| Dependencies | Nodes that must finish before execution |
| Failure policy | Continue or terminate after failure |
| Results | Structured output for callers or downstream nodes |
Create and orchestrate
- Open Configuration Management > Governance Sprites > Routines and create a Routine with a stable code, name, and description.
- Define entry parameters with types, required status, and scope.
- Add Sprite nodes with recognizable step names.
- Select each Sprite and verify inputs and permission dependencies.
- Map entry inputs or upstream outputs; check names, types, and null handling.
- Connect nodes according to dependencies. Parallelize only independent nodes.
- Configure failure policies. Critical writes and approval prerequisites generally terminate on failure; noncritical statistics may continue.
- Save and test complete paths, branches, and errors.
Execution arrows do not map parameters. Inspect each node's input source, especially arrays, objects, and empty upstream results.
Test and publish
Test normal input, empty results, missing required inputs, unavailable tools, and partial node failure. Check order, actual parameters, and business outcomes. Define confirmation or approval boundaries before publishing Sprites that change metadata, standards, quality, security, or assets.
Avoid editing definitions in active use. If supported, copy or revise the configuration, regress it, then switch consumers. Templates, space instances, and schedules are maintained separately; template publication does not overwrite existing spaces.
Schedules and execution records
Routines can be called by spaces or agents, or run through deployment-specific scheduling. Confirm frequency, enabled state, and first run in the actual environment. Inspect:
- Entry inputs and final status.
- Node start and end times, inputs, outputs, and tool calls.
- Failed nodes and errors.
- Whether Continue policies let downstream nodes consume incomplete results.
- Expected approval requests or changes from write nodes.
Diagnose nodes
| Symptom | Possible cause | Action |
|---|---|---|
| Routine does not start | Disabled schedule, empty inputs, or unpublished instance | Check instance, publication, and inputs |
| Node is skipped | Upstream termination or unmet condition | Inspect upstream state and branch conditions |
| Repeated node failure | Tool, permission, or input structure error | Run the Sprite separately and compare schemas |
| Successful node returns nothing | No matching objects or wrong output mapping | Inspect actual inputs and scope |
| Incomplete downstream results | Missing parallel aggregation or ignored failures | Add aggregation and explicit failure handling |
Relationship to other objects
Agents interpret goals and choose capabilities. Sprites perform atomic tasks. Routines define order and mappings. Spaces hold business scope, documents, and phases. Templates retain reusable configurations. Names and permissions do not substitute for one another.