Design and publish Sprites
A Sprite is a reusable atomic capability with one bounded responsibility, such as querying objects, enriching descriptions, checking standards associations, or generating proposals. Use a Routine for multiple steps, conditions, and result passing.
Plan the design
| Area | Define |
|---|---|
| Task boundary | What the Sprite handles and excludes |
| Inputs | Object identifiers, space, scope, and required parameters |
| Outputs | Text, structured results, proposals, or pending changes |
| Tools | DAM internal services, MCP, commands, or Skills |
| Permissions | Read, write, and approval requirements |
| Failures | Responses for missing input, unavailable tools, and empty results |
Avoid responsibilities as broad as governing an entire system; they make parameters, permissions, and acceptance difficult to define.
Create a Sprite
Open Configuration Management > Governance Sprites, select the agent, and create a Sprite. Enter code, name, description, and prompt using available fields. Keep the code stable; use the name for display and description for applicability and outcomes.
Input parameters
Define name, type, required status, default, and business meaning for each parameter. Prompt variables must match names. Keep object IDs, names, and display text distinct. For arrays and objects, define member structure and when empty values are allowed.
Prompt and boundaries
- State the objective and scope.
- Explain inputs and missing-input behavior.
- Specify tools and call order.
- Define output fields, format, and evidence.
- Explain how to report conflicts, insufficient permission, or uncertainty.
Enforce access in tools. Do not describe a model inference as an already completed DAM operation.
Tool binding
Depending on deployment, bind DAM tools, MCP, internal services, commands, or Skills. Use only required tools. Check address, mode, headers, timeout, and parameter JSON. A successful connection still requires output validation against a real authorized object.
Test acceptance
Test a normal object, missing required input, empty results, and tool failure. Verify mappings, actual target objects, stable output fields, and absence of unintended writes. For standards, quality, security, or asset changes, generate proposals for human review by default.
Save, enable, and publish
Verify code, schema, tools, and prompt before saving. Enabled controls availability; Expose as Tool affects discovery by agents or Routines. Record configuration, test samples, and limitations before release. Avoid overwriting configurations in active use.
Some environments support revisions, copying, ZIP import, or generation from Skills. Review generated or imported content, dependencies, and permissions before acceptance.
Troubleshooting
| Symptom | Check |
|---|---|
| Empty Sprite list | Selected agent and configuration permissions |
| Variables not substituted | Input names, prompt variables, and Routine mappings |
| Tool connection failure | Address, mode, headers, timeout, and server logs |
| Unstable output | Tighten schema, reduce unrelated tools, and define edge cases |
| Task succeeds but object is unchanged | Read-only capability and pending review |