Reusable workflows
Save procedures for work you repeat and decide when an agent should use them.
In this topic
A method stores a procedure that Mellow can find alongside skills and tools. It is useful when a task has a repeatable shape: collect inputs, call particular tools, check a condition, and return a defined result.
A method is guidance loaded into the agent's context. It should not be described as an unconditional automation engine or a guarantee that every successful conversation teaches a new routine automatically. The current method model supports user-sourced records, dependency lists, versioning, and outcome events.
Choose the right reusable component
| Component | Stores | Use it when |
|---|---|---|
| Skill | Explanations, conventions, examples, supporting references | The approach needs judgment across many tasks |
| Method | A concrete workflow body and its dependencies | A known sequence is worth reusing |
| Tool | An executable operation and its parameter schema | The agent needs to perform an action |
| Schedule | A time and instructions for an agent | The work should begin later or recur |
A method does not create a timer. A schedule does not automatically make a procedure reliable. Test the workflow before arranging repeated execution.
What a method contains
| Field | Purpose |
|---|---|
name and description | Human-readable identity and discovery text |
triggerText | Optional phrases that help identify suitable tasks |
body | The procedure, stored as text with YAML workflow support |
toolsUsed and skillsUsed | Dependencies extracted from the body |
tokenCount | Estimated context cost |
version | Revision number for edits |
createdAt and updatedAt | Record timestamps |
Write a body that distinguishes input collection, execution, and verification. Name the conditions that should stop the procedure. Avoid embedding secrets or assuming that every user has the same filesystem paths.
Discovery and loading
In automatic capability mode, the agent can search enabled capabilities with capabilities_discover and request selected entries with capabilities_load. Methods participate in that index. Loading a method also makes its referenced dependencies available through the capability-loading path, subject to the agent's allowed surface.
A method cannot grant a permission the agent lacks. If the procedure references a missing tool, repair the dependency or choose a different procedure. Do not replace an unavailable operation with an invented success result.
Outcome records and ranking
The method subsystem records loaded, succeeded, and failed events. Loading is not counted as success. The score combines observed success rate with how recently the method was used:
success rate = succeeded / (succeeded + failed)
recency weight = 1 / (1 + days since last use / 30)
score = success rate × recency weight
With no recorded outcomes, success rate is zero. This score is a ranking signal, not a certification of the method's correctness. A dependency change can invalidate a previously successful procedure.
Review a procedure after changes
After changing an API, plugin, folder structure, or model, rerun a small representative task. Inspect the actual tool output and update the method's instructions if the assumptions have changed. Retain a clear verification step so future runs can detect a partial result.
Method data uses Mellow's local storage stack. If inspecting databases outside the app, work from a safe copy and account for configured encryption. Avoid editing a live database while Mellow is using it.
See Skills for authored playbooks and Tasks for the execution lifecycle.
Continue exploring · Build your agent teamScheduled routines →Choose an agent, instructions and a time; inspect runs and missed-run behavior.