
Subtask
- 1 repo stars
- Updated February 14, 2026
- PocketConservatory/plugins
Multi-model task delegation with quota-aware routing across AI CLIs
About
subtask is a Claude Code skill in the AI & Agent Building category. Multi-model task delegation with quota-aware routing across AI CLIs
- subtask
- AI & Agent Building
- AI-coding skill
Subtask by the numbers
- Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add PocketConservatory/plugins/plugin install subtask@tonicAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| repo stars | ★ 1 |
|---|---|
| Last updated | February 14, 2026 |
| Repository | PocketConservatory/plugins ↗ |
What it does
Multi-model task delegation with quota-aware routing across AI CLIs
README.md
Tonic Plugins
A collection of Claude Code plugins for working with AI coding agents.
Quick Start
# Add the marketplace
claude plugin marketplace add PocketConservatory/plugins
# Install a plugin (e.g. subtask)
claude plugin install subtask@tonic --scope project
Restart Claude Code after installing.
Plugins
subtask — Multi-model task delegation
Route coding tasks across Claude Code, OpenAI Codex, Google Gemini CLI, and Cursor Agent based on task type and remaining quota. Never hit a quota wall on a single provider again.
subtask
Every AI coding CLI has usage limits. When you hit one, you're stuck waiting for it to reset. subtask automatically routes tasks to whichever provider has the most remaining quota, based on what the task actually needs:
| Task type | Routed to | Why |
|---|---|---|
| Code review | Codex | Separate review quota pool |
| Implementation | Codex or Claude | Whichever has more quota |
| Research / analysis | Gemini | Large context window, separate quota |
| MCP-dependent (Linear, Notion, etc.) | Claude (host) | Only Claude has MCP servers |
When your preferred provider runs out, subtask automatically falls back to the next best option — including Cursor Agent as a last resort (separate quota pool entirely).
Prerequisites
- Claude Code installed and authenticated
- At least one additional AI CLI for delegation to be useful:
- OpenAI Codex —
npm i -g @openai/codex - Google Gemini CLI —
npm i -g @google/gemini-cli - Cursor Agent — see Cursor docs
- OpenAI Codex —
- Node.js (for the quota checker,
npx cclimits)
Install
claude plugin marketplace add PocketConservatory/plugins
claude plugin install subtask@tonic --scope project
Restart Claude Code. The routing strategy and quota hooks activate automatically.
Usage
Automatic quota enforcement
subtask uses two hooks to manage quota without you doing anything:
- SessionStart — primes a quota cache (via cclimits) and injects the routing strategy
- PreToolUse — intercepts Task calls when Claude is running low on quota and directs it to use
/subtask:delegateinstead, listing available providers
Delegating tasks
Use /subtask:delegate to send a task to the optimal provider:
/subtask:delegate Research best practices for caching in Firebase
subtask will automatically pick Gemini (read-only research task, large context window, preserves Claude quota).
You can also force a specific provider:
/subtask:delegate Review the auth changes using codex
/subtask:delegate Use gemini to analyze the performance impact
/subtask:delegate Implement retry logic with codex
Routing
Priority: Claude → Codex → Gemini. Claude is the primary model. External CLIs are overflow buckets when Claude is under pressure.
The PreToolUse hook computes pressure levels from remaining quota and reset times. When Claude is HIGH or CRITICAL and alternative providers are available, it blocks Task calls with a directive to delegate. Code reviews always go to Codex (separate quota pool).
Skills
| Skill | Description |
|---|---|
/subtask:delegate <task> |
Route a task to the optimal AI CLI based on type and quota |
License
MIT