Start a session safely

Goal

Before the first write of a session, confirm which repository is open, which modeling technologies are really loaded and enabled, and exactly what will change. For an architect, or anyone about to modify a model. You end with a short, evidence-based statement of the session's starting state, an agreed list of intended writes, and the right skill chosen for the job.

Where it runs

This scenario needs the MCP server. Computer use covers its UI steps.

Before you start

• The plugin installed and verified (Install and verify).
• EA running with the intended model open, or the path to it.
• A statement from you of what you are trying to do. The routing to the next skill depends on it.

Most Enterprise Architect failures are not modeling errors. They are sessions acting on a model that is not the one assumed, on a technology that is installed but not active, or on a repository different from the one in mind. The check is four reads and a conversation. It costs little and removes the most expensive mistakes. The skill that carries it, ea-start-here, also routes the task to the next skill, so it is the right first message of any session that touches an EA model.

Steps

1. AI Power Tools ea-start-here confirms the server is alive and notes its version (ping).
2. AI Power Tools ea-start-here reads the open repository (get_repository_info). If EA is not running, the AI says so and stops, then offers to start it, through the server's launch operation or with computer use, after you agree. Starting EA and clearing any dialog it raises is a UI step: AI with computer use; otherwise you. If a different model is open, it stops and asks which is intended.
3. You Settle which model is the target, if step 2 did not make it obvious.
4. AI Power Tools ea-start-here and ea-mdg-assess classify the modeling-language state (assess_mdg_situation), a rule-based reading of the stereotypes in use and the technologies loaded. A technology disabled in Manage Technology does not count as loaded in this reading.
5. AI Power Tools ea-start-here checks a specific technology (get_mdg_from_runtime, which needs the technology id, such as WBA, not its display name). As needed it also lists registered technologies and the technologies stored in the model, each with its Location.
6. You Agree the intended writes and the target repository before the first write. The AI states them in plain terms: "I am about to create about 120 elements and 140 connectors in <model-dir>\WestbrookBank.qea, under a new package. Nothing existing will be changed. Agree?"
7. AI ea-start-here routes the task to the right skill. Confirm the choice if the task is ambiguous.
8. You Accept or decline a baseline before a risky change. ea-change-management takes it. See Protect a change with a baseline.

The check stops at the first read that fails. Do not skip ahead.

Reading the classification. assess_mdg_situation returns a scenario number, a plain-language description, the loaded technologies, the Sparx-shipped languages loaded, the count of ad hoc stereotypes, the percentage of elements covered, and a suggested next skill.

This is a rule-based reading: treat the description as the finding and the suggested skill as a suggestion. Check that the skill it names appears in the routing table of ea-start-here before acting on it. Assess the language situation goes deeper.

Three different questions about a technology. A mismatch among these is usually the problem.

Installed is not active, and loaded is not enabled: EA reports a technology as loaded even when it is disabled in Manage Technology, so the runtime reading reports a disabled technology as registered but not enabled. Omitting the technology id returns an error, not a list.

EA keeps a technology that is imported into a model at one of two Locations, which Manage Technology shows: Project or Model. A Model copy holds the whole technology, toolbox pages included. A Project copy holds the profiles and diagram types and no designed toolbox pages. If the model holds the same technology id at both, EA answers from the Project copy, and the runtime reading says so. The modeling-language part of Governance covers the two Locations.

Prompts to use

Opening

Follow-ups

Asks the AI to read the description and check the recommendation against the routing table.

Moves to Protect a change with a baseline.

Recommended models

Ratings are guidance based on the kind of reasoning the skill needs, not measured benchmarks.

Checking the result

• Accurate. The repository the AI names is the one open in EA: compare the path to EA's title bar, and re-read it if there is any doubt. The loaded technologies match what you see in EA's Manage Technologies dialog, with "Hide disabled Technologies" unticked, including each entry's Location. A technology you see disabled there reads as registered but not enabled. The AI can open the dialog with computer use; otherwise you open it.
• Complete. The summary states the repository, the EA version, the scenario with its reason, the loaded technologies with their source, the intended writes, and the skill chosen.
• Nothing unintended changed. The check only reads. If a baseline was taken, the baseline list shows it, and nothing else in the model differs.

Variations

• A stereotype that behaves inconsistently. Run steps 4 and 5 first. An element can carry a stereotype from a technology that is registered but not loaded, or loaded but disabled in Manage Technology.
• Before any MDG task. This check is the entry to Assess the language situation.
• After a restart or a clear of the conversation. Re-read the repository and answer questions about the model by asking it, not from memory of earlier steps. Elements move, get renamed and get deleted.
• On a new machine. Run this directly after Install and verify.

Limitations

• The assessment is a rule-based reading. It reads stereotype usage and the loaded languages. It cannot see a profile package in the model that could be the source of a technology. If you know a source model exists, build from the source model regardless of the suggestion.
• The state checks can disagree with EA's own dialog. EA's automation interface does not list technologies, so a technology that exists only as a disabled file on the machine, with no copy in the model, may not appear in the registered-technology list. Cross-check in Manage Technologies when it matters.
• The list of available modeling languages is not exhaustive. EA offers no way to list technologies through automation. AI Power Tools combines fixed tables with a probe list.
• Setting an active technology records a preference only. It enforces nothing. Use a ruleset audit (ea-validation) to enforce conformance.
• If a request seems to hang, look at EA. A modal dialog blocks the connection and names its cause. With computer use, the AI sees the dialog in EA's window, reads it, uses the message to correct the call, dismisses it and carries on. Without computer use, you take a screenshot of the dialog and paste it into the conversation so the AI can analyze it, then dismiss the dialog so processing can continue. See Troubleshooting.