mcpField
A
autodiscovery

autodiscovery

github.com/allenaiData & databases4267 installsUpdated 1d ago
Repository

About autodiscovery

A routing skill for AutoDiscovery, an AI-driven scientific discovery platform that iterates experiments using Bayesian surprise and MCTS optimization. It directs users to the hosted Asta service or local CLI workflow based on their request.

Key features

  1. 01Routes requests to remote or local workflows
  2. 02Uses explicit wording to select a workflow
  3. 03Requires reading workflow instructions first

Use cases

  1. 01Run experiments through the hosted Asta service
  2. 02Run experiments locally with the CLI

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/allenai/asta-plugins --skill autodiscovery

After installing, type/autodiscoveryin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

assets/
interpreting-results.md
workflows/
local.md
remote.md

Details

From the project README

AutoDiscovery

AutoDiscovery is an AI-driven scientific discovery platform that runs iterative experiments guided by Bayesian surprise and MCTS optimization.

This skill is a router. Inspect the user's request, pick one workflow, then read its .md file in workflows/ and follow it. Do not execute a workflow from memory — always open the file first.

Workflows

NamePurposeDetailed instructions
remoteUse the hosted AutoDiscovery service via asta autodiscovery. Requires Asta auth and consumes credits; runs execute on Asta infrastructure.workflows/remote.md
localRun AutoDiscovery locally via the asta-autodiscovery CLI. Executes on the user's machine with user-provided API keys for making LLM callsworkflows/local.md

Routing

1. Honor explicit requests
  • "local", "on my machine", "offline", "run it myself", or any reference to the asta-autodiscovery CLI → local.
  • "remote", "hosted", "cloud", "Asta service", "my runs", "credits", or any reference to the asta autodiscovery (note the space) subcommand → remote.
2. Use signals when the request is ambiguous
  • Past runs are stored in a local directory for local runs, or on the remote server for remote runs
  • If the user is starting fresh and has not indicated a preference, ask which workflow they want before proceeding. Mention the tradeoff: remote uses Asta credentials and rate limits; local runs with the user's hardware and credentials.
3. No chaining

The two workflows are independent. Do not silently switch between them mid-task.

Comments

Markdown supported