luisdomarco Entity Selector Skill
Applies the entity decision tree to select the correct type for each responsibility identified in a process. Avoids intuition-based selection and ensures architectural coherence.
Input / Output
Input:
- Description of a responsibility or capability to model
- Full process context (to evaluate relationships with other entities)
Output:
- Recommended entity type with justification
- Signals that confirm the selection
- Alerts if there is ambiguity between two types
Procedure
1. Extraction of Structural Metrics
The decision trees, discriminatory matrices by shared attributes, and relational anti-pattern policies no longer reside hardcoded here. You avoid biases by getting the updated "Architecture Component Metrics" from the centralization framework.
You must first and mandatorily read the following resource before activating the deductive logic:
../../resources/res-architecture-component-metrics.md
Extract from there blocks 1 through 4 (Decision Tree, Discriminatory Threshold, Edge Case Resolution, and Prohibited Anti-Patterns) and use them to resolve the structural question.
5. Selection output
For each analyzed responsibility, deliver:
Responsibility: [description]
Selected entity: [TYPE]
Justification: [why this type and not another]
Confirmatory signals:
- [signal 1]
- [signal 2]
Alerts: [if there is residual ambiguity or risk of anti-pattern]
Examples
Example 1 — Correct selection of Skill vs Agent
Responsibility: "Format the output as structured JSON according to a fixed schema."
Analysis:
- Does it condition behavior? No.
- Is it static information? No.
- Is it manual/deterministic? No.
- Multiple responsibilities? No.
- Needs own identity and decision? No — always does the same given the input.
- Reusable by multiple agents? Yes.
→ SKILL (ski-format-json-output)
Example 2 — Detection of required Workflow
Responsibility: "Receive an email, classify it, look it up in the CRM, generate a response and send it to the client."
Analysis:
- Multiple responsibilities? Yes: classify, look up in CRM, generate response, send.
- Transfer of outputs between parts? Yes: classification feeds the lookup, which feeds the generation.
→ WORKFLOW + Agent Specialists for each differentiated responsibility.
Example 3 — Hook vs Rule selection
Responsibility: "Run linter automatically every time a file is written."
Analysis:
- Does it fire on a system event? Yes — PostToolUse:Write.
- Does it execute something? Yes — it runs a linter script.
- Is it event-driven? Yes — fires automatically without manual invocation.
→ HOOK (hok-lint-on-write) for the event trigger.
Not a Rule: Rules don't execute or delegate to scripts — they passively constrain behavior.
Note: Avoid using type: "prompt" hooks on high-frequency events — they interrupt every operation. Reserve prompt hooks for low-frequency events (SessionStart, Stop).
Error Handling
- Unresolvable ambiguity: If after applying the tree and the table the selection is not clear, present both options to the user with their justification and ask them to decide.
- User proposes an incorrect type: Explain the anti-pattern it would generate and propose the correct alternative with justification.