Works with the AI Agents you already use
EcoRun
Picks the right Claude model (Haiku, Sonnet, Opus or Fable) and effort level for each task, so you spend fewer tokens without losing quality.
Free
EcoRun
Example session with this skill installed
Migrate this whole React app to TypeScript, then pull the title, author and date out of the 50 PDFs in /docs into a CSV.
- Read your context and instructions
- Compiled the ecorun
EcoRun — before we start
Task: large refactor (React → TypeScript) + bulk data extraction from 50 PDFs.
Suggested model: Fable · medium (if you don't have it, Opus · high) for the migration. You're currently on Sonnet.
Why: a full migration needs architecture-level judgment. A smaller model tends to miss cross-file type issues, and redoing the work costs more than doing it right the first time.
To continue:
- Switch the model with
/model fableand reply "continue". - Or reply "stay here" to do it with Sonnet.
What happens after you choose
| Step | Who does it | Why |
|---|---|---|
| Plan and run the TypeScript migration | Main model (Fable or Opus) | Needs judgment; edits stay in the main model |
| Find every component, prop type and import to update | eco-explorer subagent (Haiku) | Searching is cheap and mistakes are easy to spot |
| Extract title, author and date from the 50 PDFs into a CSV | eco-extractor subagent (Haiku) | Repetitive, high-volume, easy to verify |
| Check the CSV against a sample and run the build | Main model | Verification before calling it done |
EcoRun stops only on substantial tasks. Short questions get answered right away, and once you choose a model it won't ask again for the same task.
Connects securely to your tools. The creator never sees your data.
About this skill
Using the biggest model for everything is comfortable but expensive. Many tasks (searching files, reading logs, extracting data, applying a template) are done just as well by a smaller, faster model. And the opposite is also true: dropping to a small model on a hard task costs more, because you end up redoing it.
EcoRun applies Anthropic's official model-selection guidance to every task:
- Haiku: searching, reading logs, extracting, classifying (mistakes are easy to spot)
- Sonnet: well-scoped work with a clear spec, everyday documents
- Opus: work that needs judgment, design or hard debugging
- Fable: the hardest, longest tasks (falls back to Opus at high effort if you don't have access)
How it works:
- Before a substantial task, EcoRun classifies it. If another model fits better, it stops, suggests the model and effort level, and waits for you to choose, instead of spending tokens first and recommending later.
- It never stops on short questions, never asks twice for the same task, and stops asking if you tell it to.
- The golden rule: when in doubt between saving and quality, quality wins.
Works in Claude.ai, Cowork and Claude Code. EcoRun is built for Claude models. It responds in your language.
Claude Code users: install the full plugin, which adds three Haiku subagents (explorer, log reader, extractor) where most of the savings come from: /plugin marketplace add jorge2510/EcoRun /plugin install ecorun@ecorun
Source, docs and updates: https://github.com/jorge2510/EcoRun License: MIT + Commons Clause. Free to use and modify, including at work. Reselling requires the author's permission.
Requires user to manually change models in UI. Best for tasks long enough to offset the cost of the gatekeeping turn. Cannot force model changes via API without implementation in your own code.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
This skill has been used 1 time.
Trust & safety
Security scan
Security scan passed 6 days ago
- Free with an Agensi account
Permissions
No special permissions declared or detected