LLM context files

You can download the file here:

prompt_context.zip  

prompt_context.zip is a self-contained context package for building OKMFlow workflows with an AI assistant (Claude, ChatGPT, or similar). It bundles the workflow engine reference, the OpenKM integration reference, a set of documented training examples, and a set of ready-to-use workflow templates, so the assistant does not need any other source of information to generate a working .okmflow file.

What's included

• Reference files (llms-full-*.txt, llms-small-*.txt) ? the OKMFlow engine (nodes, Groovy scripting, WorkflowUtils, workflow database schema), the workflow forms XML format, and the OpenKM server integration layer (internal Java API, REST SDK, plugins, metadata, core utilities, repository database schema).
• workflow-form-x.x.dtd ? the formal grammar for workflow task forms.
• Training samples (numbered folders, e.g. 01-quick-start-guide/ ? 14-Rest request sample/) ? small, focused .okmflow examples, each one illustrating a single feature (variables, task assignment, transitions, mail, libraries, error handling, REST calls, etc.), together with their diagrams.
• Ready-to-use templates (template-*/ folders) ? complete, deployable workflows for common business scenarios (vacation request, purchase order, expense report approval, complaint management, shared global library). Use one of these as a starting point when your scenario is close to it, instead of starting from an empty skeleton.
• INDEX.md ? describes every file in the package in detail, including exact file names and versions. Point the assistant to it if it needs to locate a specific reference file.

How to use it

1. Start a new conversation with your AI assistant and upload prompt_context.zip.
2. Tell the assistant to use the zip as context for the whole conversation.
3. Describe the workflow you want in plain language: the steps involved, who does what, and the outcome.
4. Ask the assistant to generate the .okmflow file.
5. Import the generated file into the OKMFlow designer and run it to check that it behaves as expected.

Minimal example

A single prompt with the minimum information is usually enough for a first draft:

I have uploaded a zip file with the OpenKM workflow engine documentation. Use it as context for everything we do in this conversation.

Build a vacation request workflow: an employee requests a date range, a manager approves or denies it, and the employee receives an email with the result.

Please generate the .okmflow file.

Building it in stages

For workflows with several roles, loops, or conditional routing, it generally works better to build the workflow incrementally instead of asking for everything at once:

1. Skeleton ? ask only for the nodes and transitions, no scripts or forms yet. Import it, arrange the nodes visually, and upload the corrected file back so the assistant keeps that layout in the following steps.
2. Routing and logic ? describe any Decision nodes, loops, or Action nodes needed to control the flow.
3. Forms ? define the fields for each task, which ones are editable versus read-only, and how values are pre-filled between tasks (the data attribute).
4. Task assignment ? describe who each task should be assigned to (a fixed user, the initiator, a role pool, or the previous actor on a loop-back).
5. Notifications ? add the Mail nodes and describe what each email should contain.
6. Testing and fixes ? run the workflow, and paste back any error message from the OpenKM log together with the file. The assistant can usually diagnose and fix it directly.