ml-autoresearch - Analysis-first autonomous ML research loop
Runs an autonomous ML research loop that analyzes each experiment, applies one evidence-based change, and optionally grounds changes in scientific literature.
Tags
Updated: 2026-10-08Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Resolve experiment bindings
- Run one experiment per iteration
- Analyze model behavior
- Ground changes in literature
- Record experiment results
- Keep or revert changes
- Continue until interrupted
Inputs
- Project source files
- Training configuration
- Run command
- Scalar metric
- Metric direction
- Editable file list
- Iteration strategy
- Time or epoch limit
- Literature setting
- Research domain
- User confirmations
Outputs
- Resolved loop.run.yaml
- Experiment results ledger
- Analysis plan files
- Analysis result artifacts
- Code snapshots
- Git branches or commits
- Literature findings ledger
- Updated experiment state
Requirements
- Python 3.9 or later
- Write access to the working directory
- A runnable ML experiment
- A measurable scalar metric
- literature-search skill when literature is enabled