version: "1.0.0"

Evaluation Creation Dispatcher

Select the correct evaluation-generation flow by matching the user's requested scenario to the configured agent's topics. This skill only routes; it never generates, edits, or pushes evaluation files itself.

Rules

• Route by whether the requested scenario is implemented by a configured topic,

not by whether the user knows or mentions internal topic terminology.

• Inspect topic evidence before choosing a generation skill.
• Route new generation to exactly one generation skill. Existing-set edit/add

requests go to the update skill instead; never regenerate to reach editing.

• Do not expose skill names, file paths, or internal routing terminology to the

user.

• After selecting a flow, read the selected skill and follow it completely.

Step 1: Identify the requested scenario

First recognize an existing-set edit/add request, including a response to Edit this existing set or Add test cases to this set. If the user already selected an exact folder/source, carry it and the requested action directly to src/skills/evaluations/update/SKILL.md. Otherwise use that skill's evaluation_review.py --list-all discovery and explicit selection before any mutation. Do not ask for a new scenario or pick a same-name set from another source. For all other requests, continue below.

Extract the scenario, goal, or capability the user wants to evaluate, such as compensation, employee ID, HR policy lookup, or IT ticketing.

If the request does not identify a scenario or goal, ask:

Which scenario or goal should I create evaluation tests for?

Wait for the answer before routing.

Step 2: Search configured topics

Read .local/config.json when it exists. A configured topic inventory is available only when:

1. setup is "complete";
2. agent.folder is present; and
3. {agent.folder}/topics/ contains at least one .mcs.yml file.

When the inventory is available, inspect every topic's:

• Filename and topic display name.
• modelDescription.
• triggerQueries.
• User-facing messages and invoked actions when needed to confirm behavior.

Match by capability and meaning, not only exact words. For example, a compensation request may match a topic described as base pay or salary.

Step 3: Route from the match

Matching topic found

When one or more topics clearly implement the requested scenario, read src/skills/evaluations/create/SKILL.md and follow it. Carry the requested scenario and matched topic names into that skill as the generation scope.

If several topics contribute to the same scenario, include all of them. Do not ask the user to select topics unless the candidates represent genuinely different scenarios.

The selected generator owns existing-set collision choices and renders the pre-generation coverage explanation from src/skills/evaluations/experience-contract.md once. Do not duplicate it here.

No matching topic found

When no configured topic implements the requested scenario, or no configured topic inventory is available, read src/skills/evaluations/generate/SKILL.md and follow it.

The catalogue-grounded result is still a valid .mcs.yml plus CSV evaluation set, but it represents the requested ESS scenario rather than behavior observed in the configured agent.