Claude Skill MartBrowse skillsQuick linesLearn by videoTerminal guideWhat is a Skill?
 
← Back to list

Agent Routing (model, effort & cascade selection)

Decides which model, effort level, and cascade shape each subagent gets, routing on measured cost per completed task instead of per-token price.

AutomationAdvanced★ 150⑂ 6AI score 9/10Last updated: Oct 2, 2026

What it does

  • Gives a task-shape routing table: two questions (short vs. long output, mechanically checkable vs. judgment) map onto haiku@low, sonnet@medium, opus@high, and so on.
  • Routes on measured cost per completed task, not per-token price — with data showing a 5×-cheaper tier costing 30% more per solved task because it emitted 6.7× the tokens.
  • Supplies cascade design rules: check the cheap rung is actually cheaper first, no verifier ⇒ no cascade, and the verifier's holder (the orchestrator) makes the escalation call because workers self-report success even when they failed.
  • Documents the concision lever (−37% Sonnet / −27% Haiku output on long generation) and where it stops helping, plus how thinking suppression halves pass rates.
  • A context handoff checklist: subagents inherit nothing, so artifact paths, verbatim commands (with interpreter path), anti-patterns, and an output spec must be serialized into every spawn prompt.
  • Loop discipline (out-of-band evaluator, argmax selection, stop on first regression) and a procedure for watching a subagent fan-out live via per-thread streams.

Who it's for

  • Claude Code / Claude Code on the Web users spawning subagents through Agent or Workflow tools
  • Teams whose agent pipelines cost more than expected and who want to know why
  • Orchestration designers who want an evidence-based tier policy rather than "just use the big model"
  • (Not applicable to plain claude.ai chat usage)

Example uses

  1. 100-way JSON extraction fan-out: short, checkable output → haiku@low plus schema validation, with an explicit argument against up-tiering "to be safe" (3–5× cost, no measured gain).
  2. Spec-to-module generation: rung 1 sonnet@low + concision, and on test failure retry the same model at medium carrying the prior patch and raw test output — cheaper and more accurate than jumping to opus.
  3. Four explore agents on a 2,300-file repo: hand each one an index slice, the exact command, and a no-ls/glob rule so discovery cost drops to roughly zero.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/oaustegard/claude-skills/HEAD/agent-routing/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the agent-routing folder from the GitHub repo oaustegard/claude-skills into my ~/.claude/skills/agent-routing/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/oaustegard/claude-skills.git && mkdir -p ~/.claude/skills && cp -r claude-skills/agent-routing ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal and create the skills directory: mkdir -p ~/.claude/skills
  2. Clone the repository: git clone https://github.com/oaustegard/claude-skills.git
  3. Copy just this skill: cp -r claude-skills/agent-routing ~/.claude/skills/
  4. Confirm both ~/.claude/skills/agent-routing/SKILL.md and its references/ folder came along — the reference files hold the calibration data.
  5. Restart Claude Code and check that agent-routing shows up in your skills list.
  6. Trigger it by asking "which model and effort should this task get?" or whenever you ask Claude to fan out several subagents.
  7. Note that the price and token figures are snapshot measurements; if your models or pricing have changed, re-measure on your own workload instead of trusting the tables as-is.
View source on GitHub ↗License: MIT