arthai-marketplace

/routing

Re-show the triage routing table and SPEED scoring rubric on demand — decide toolkit vs built-in for the task at hand.

Synopsis

/routing <task description>

When to use it

• You want the full routing rubric back mid-session — the triage-router hook shows it once, then only compact reminders
• You’re unsure whether a task warrants a heavyweight workflow or a quick built-in edit, and want it scored
• You’re learning the toolkit and want to understand why a request routes where it does
• Not for: listing what’s installed — that’s /skills, which /routing calls to enumerate live targets

Quickstart

/routing

What you’ll see: the SPEED scoring rubric (the five axes used to decide toolkit-vs-built-in) followed by the live catalog of installed routing targets.

Examples

/routing                       # show the SPEED rubric + installed routing targets
/routing fix the auth bug      # score this task and recommend a route
/routing rename a variable     # low SPEED score (0-1) → recommends built-in Edit, not a workflow

Arguments & flags

Argument	Values	Default	What it does
task description	free text	none	When present, the task is scored 0–5 on SPEED and a concrete route is recommended

What it does

1. Prints the SPEED scoring rubric — Scope, Project-context, Expertise, Effort, Dollars (1 point each).
2. Runs /skills to enumerate the routing targets actually installed, so it never recommends a command you don’t have.
3. If you passed a task description, scores it, shows the breakdown, and ends with a one-line recommendation: → route: <skill/agent/tool> (SPEED=<n>).

SPEED scoring rubric

Score the task on five axes, 1 point each:

Axis	Question	+1 when
S — Scope	How many files/steps?	3+ files or steps
P — Project	Needs CLAUDE.md / project context?	Yes
E — Expertise	Domain-specific (QA/SRE/design/architecture)?	Yes
E — Effort	Multi-turn workflow (PR/deploy/planning)?	Yes
D — Dollars	Would a cheaper model (Haiku/Sonnet) suffice?	Yes

Decision rules

• 0–1 → Claude Code built-in tools (Read / Grep / Glob / Edit / Bash).
• 2+ → a toolkit skill (multi-step workflow) or agent (domain expertise + project context).
• 3+ and multi-layer → /planning → /implement (team orchestration).
• 3+ and risky/sensitive → architect (Opus) brief → executor agent (Sonnet).

Output & artifacts

• The rubric, catalog, and (if a task was given) a routing recommendation, printed into the conversation. Nothing is written to disk; no agents are spawned.

Troubleshooting

Problem	Fix
Routing targets look wrong/missing	The catalog comes from /skills; if a command is absent, install the bundle that provides it
Recommendation feels too heavy for a tiny change	See Decision rules — SPEED 0–1 routes to built-in tools; re-score, since a 3+ score means the task genuinely spans files/context/expertise
Table doesn’t appear automatically anymore	That’s expected — the hook injects it once per session to save tokens; /routing is the manual re-show

Related

• /skills — the installed catalog this rubric routes to
• /onboard — project briefing and prioritized next steps
• Workflow comparison — deeper decision tree for the autonomous/team workflows