Skill role taxonomy

Defines bounded architecture roles for routing, retrieval, transformation, reasoning, verification, policy, memory maintenance, and structured output without claiming that a compatible model exists.

Research
Last verified
Not verified
Updated
Reading time
2 minutes

Defines bounded architecture roles for routing, retrieval, transformation, reasoning, verification, policy, memory maintenance, and structured output without claiming that a compatible model exists.

Architecture guide: this topic defines a modular tiny-model planning contract. It does not claim that model artifacts exist, are compatible, or execute on this WordPress site.

Role before model

Start by naming the bounded function: route, retrieve, transform, reason, verify, constrain, maintain memory, or emit structure. Then define what evidence would show that a candidate model satisfies the role.

Minimum role contract

  • Accepted input schema and maximum size.
  • Produced output schema and confidence semantics.
  • Supported domain and explicit out-of-domain behavior.
  • Quality metrics and failure thresholds.
  • Resource profile and concurrency limits.
  • Data, privacy, policy, and authority boundaries.

Anti-pattern

Do not label a general model “legal,” “medical,” “security,” or another high-stakes role based only on fine-tuning claims. The role requires task-specific evaluation, limitations, and governance.

Scope

This starter page defines the questions, boundaries, evidence, and failure modes that should be recorded before a capability is presented as supported.

Engineering considerations

  • Identify the source, version, target environment, and owner.
  • Separate observed values from estimates and externally reported values.
  • Record trade-offs, unsupported cases, and fallback behavior.
  • Link performance statements to a compatible benchmark methodology.

Verification questions

  • What exact artifact, revision, backend, and environment were reviewed?
  • Which assumptions could change the result?
  • Which data should be retained so another engineer can reproduce the conclusion?