From mk
Classifies task complexity by matching domain keywords against a CSV, returning complexity level, workflow intensity, and model tier override for Phase 0 orchestration.
How this skill is triggered — by the user, by Claude, or both
Slash command
/mk:scale-routingWhen to use
Auto-invoked at Phase 0 — scans task description for domain keywords, returns complexity tier and model override. Not user-callable directly.
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Automatically classifies task complexity based on domain keywords.
Automatically classifies task complexity based on domain keywords.
Internal sub-skill of Phase 0 (Orient). Not user-invocable directly — invoked by the orchestrator during task classification.
Replace subjective orchestrator judgment with deterministic, auditable domain-based routing. A fintech auth change should auto-classify as COMPLEX without human intervention.
data/domain-complexity.csvreferences/multi-layer-detection.md)bug_fix, feature, refactor, security, devops, docs, review, intake (load references/task-type-classification.md).claude/product-areas.yaml if exists — merge area keywords and paths (see references/product-area-config.md){domain, level, workflow, model_tier_override, task_type, suggested_skill, confidence, product_area?}unknown, defer to manual classification| Column | Values | Purpose |
|---|---|---|
| domain | snake_case identifier | Domain name |
| signals | comma-separated keywords | Match against task description |
| level | low, medium, high | Complexity classification |
| workflow | one-shot, standard, enhanced, advanced | Workflow intensity |
| web_searches | comma-separated topics | Suggested research queries |
| Level | Model Tier | Gate 1 | Workflow |
|---|---|---|---|
| low | TRIVIAL (Haiku) | Bypass eligible (one-shot) | Minimal |
| medium | STANDARD (Sonnet) | Required | Standard phases |
| high | COMPLEX (Opus) | Required | Full phases + security |
Called automatically by orchestrator at Phase 0. Not invoked directly by users.
Orchestrator Phase 0:
1. Run mk:scale-routing on task description
2. If match found → use returned level/workflow/model_tier
3. If no match → fall back to manual classification per model-selection-rules.md
Users can add custom domains by editing data/domain-complexity.csv. Add a new row with domain keywords relevant to your project.
mk:autobuild/scripts/density-select.sh — if the autobuild workflow skill is removed or renamed, density selection breaks. The script is shared intentionally (single source of truth for density policy); treat it as load-bearing infrastructure, not a private harness internal.Base fields (v1.0):
| Field | Values | Description |
|---|---|---|
domain | snake_case string | Matched domain or unknown |
level | low, medium, high | Complexity classification |
workflow | one-shot, standard, enhanced, advanced | Workflow intensity |
model_tier_override | TRIVIAL, STANDARD, COMPLEX | Forced model tier |
New fields (v2.0):
| Field | Values | Description |
|---|---|---|
task_type | bug_fix, feature, refactor, security, devops, docs, review, intake | Classified task type |
suggested_skill | mk:fix, mk:cook, mk:cso, mk:review, mk:intake | Recommended skill for this task |
confidence | HIGH, MEDIUM, LOW | Routing confidence from Layer 3 scoring |
product_area | string | Area name from .claude/product-areas.yaml (omitted if no YAML) |
New fields (v2.1 — Phase 5 of harness plan, 260408):
| Field | Values | Description |
|---|---|---|
autobuild_density | MINIMAL, FULL, LEAN | Recommended scaffolding density for mk:autobuild runs (see Adaptive Density Policy below) |
Used by mk:autobuild to choose how much scaffolding to apply per run. The decision rules:
level (model_tier) | model id contains | autobuild_density |
|---|---|---|
| low (TRIVIAL/Haiku) | any | MINIMAL |
| medium (STANDARD/Sonnet) | any | FULL |
| high (COMPLEX/Opus) | opus-4-6 or opus-4.6 or opus-4-7 | LEAN |
| high (COMPLEX/Opus) | other (e.g., opus-4-5, claude-opus-4) | FULL |
Override: MEOWKIT_AUTOBUILD_MODE=MINIMAL\|FULL\|LEAN env var, when set, overrides the auto-detected value. The override is logged in the autobuild run report for audit.
For scriptable density selection, callers may invoke .claude/skills/autobuild/scripts/density-select.sh which echoes only the density token to stdout.
data/domain-complexity.csv — Domain → complexity → workflow mappingreferences/multi-layer-detection.md — 4-layer detection logic and confidence scoringreferences/task-type-classification.md — 8 task types, signals, and suggested skillsreferences/product-area-config.md — Optional .claude/product-areas.yaml schema and loading rules.claude/skills/autobuild/scripts/density-select.sh — Echoes autobuild_density token to stdout; used by autobuild when scriptable density selection is needednpx claudepluginhub ngocsangyem/meowkit --plugin mkScores task complexity 1-10 across 8 dimensions, selects skill chains (BUG/FEATURE/REFACTOR/ARCHITECT), assigns model tiers, searches past patterns before non-trivial tasks.
Classifies tasks (7 types/4 sizes), routes S/M via direct tools and L/XL to code-researcher for multi-package investigation, and produces fully coded plans for execution.