Repeatable work
Skills and memory
Keep the useful parts of a completed task: the method to repeat, the context to remember, and the evidence that should remain in its source system.
A skill captures how to do the work
A recurring report usually needs more than a prompt. It has source definitions, matching rules, exclusions, a preferred output, and checks that prevent familiar mistakes. A Proto skill packages those instructions so the method can be used again.
Skills are files. A skill directory contains a SKILL.md instruction file and can include supporting references, scripts, or assets. Workspace skills live with the project; global skills are available across workspaces. Proto can help write or revise a skill after a successful task.
Built-in skills cover areas such as business files, document creation, browser work, ERP•AI apps, and workflows. A custom skill adds the context specific to your process. It should describe the conditions in which the instructions apply, not assume that every future task is identical.
Begin with a method you have reviewed
Capture a process after you understand its inputs and failure cases. For a weekly operations report, useful instructions include the reporting cutoff, the meaning of each metric, the matching keys, and what to do when a source is missing.
Create a workspace skill for our weekly supplier review. Use the reporting definitions and workbook structure from this completed task. Include checks for duplicate supplier IDs, mixed currencies, and missing delivery dates. Require unresolved data issues to appear in the review notes.
The skill should help someone inspect the next result. Include examples of valid input and exceptions when they clarify the work. Keep credentials and changing facts out of the instructions; reference the source that owns them instead.
Memory captures useful context
Memory serves a different purpose. It retains workspace context that may help later tasks: terminology, project decisions, preferences, or pointers to important material. Proto stores this context in a local memory index and topic files under .proto/memory/.
Memory is selective context, not a complete or authoritative record of the business. Model context has limits, and a remembered fact may become stale. Keep contractual terms, financial balances, approvals, and operating records in their source systems. Ask Proto to revisit those sources when freshness matters.
| Keep in | Suitable material |
|---|---|
| A skill | A repeatable method, output requirements, and review checks |
| Workspace memory | Stable terminology, useful decisions, and source pointers |
| The source system | Current records, accountable approvals, and official documents |
Review changes like other working files
Because skills and memory are local files, teams can inspect their contents and manage them with their normal file or version-control practices. Assign an owner to a shared method. When a metric definition or business rule changes, update its instructions and test them against a representative example.
A skill can include executable scripts, so installing one is a trust decision about its contents and dependencies. Read what it asks Proto to do before using it with important data or privileged connections.
Use workflows for event-driven execution
A saved skill does not by itself schedule work or listen for new records. It guides a task when used. If a business event or schedule should start a repeatable process in ERP•AI, use a workflow with an explicit trigger and tested actions.
Skills, memory, and workflows can support the same process, but each has a distinct job. Keeping those jobs clear makes the process easier to review, update, and hand to another person.