| name | exeris-task-classifier |
| description | Router/Planner triage skill for Exeris Kernel. Classifies task type, scope, severity, and recommends primary agent based on primary risk. |
| argument-hint | Task description + changed files/modules (if known) + expected outcome |
| user-invocable | true |
| disable-model-invocation | false |
Exeris Task Classifier
Purpose
Classify incoming work before execution starts.
This is a triage skill: it does not implement changes. It identifies task class, scope, and likely ownership.
Output Contract
Return exactly:
task_class (ARCHITECTURE | IMPLEMENTATION | CONTRACT_VERIFICATION | PERFORMANCE_REVIEW | DOCS_ADR | MULTI_DOMAIN)
scope (single-module | cross-module | single-subsystem | cross-subsystem)
severity (low | medium | high | critical)
primary_risk
recommended_primary_agent
Classification Heuristics
ARCHITECTURE: boundaries, module placement, dependency direction, ADR alignment.
IMPLEMENTATION: code change with stable boundaries and contracts.
CONTRACT_VERIFICATION: observable behavior or SPI/TCK-sensitive behavior.
PERFORMANCE_REVIEW: hot-path allocation/copy/concurrency risks.
DOCS_ADR: docs contract drift, ADR updates, subsystem doc synchronization.
MULTI_DOMAIN: at least two classes above are first-order concerns.
Guardrails
- Preserve The Wall in every classification.
- If uncertain between two classes, emit
MULTI_DOMAIN and state both dominant concerns.
- Prefer the smallest sufficient docs set (
docs/modules, docs/subsystems, then ADRs if needed).
Completion Criteria
Classification is complete only if all five output fields are present and justified in 1-2 concise bullets.