/ml-ablation

Make it your own.

In Claude Code, use the slash command and add your context. In Codex, select ml-ablation from the just-vibe skill picker, then send the same brief.

Version 0.11.0 also supports /jv ml-ablation, /just-vibe ml-ablation and /jv:ml-ablation in Claude. See shortcut setup and context examples.

What the agent does

1. State the causal comparison: one meaningful variation at a time or a justified factorial design.
2. Hold data and evaluation protocol constant, and repeat seeds or matched runs where variance could overwhelm the effect.
3. Compare each variant's effect with its uncertainty and cost.

Inputs

• hypothesis, reference model, feature/component variants, fixed evaluation, and budget.

Optional context: scope, references, constraints, successCriteria, environment, mode, budget.

Scope

Reads

Isolate the contribution of specified components using controlled comparisons.

Writes

Inspect/plan: inspect or propose; save requested artifacts only. Apply: make the requested changes or execute the requested operation within its resolved target and limits. Local preparation does not authorize live, remote, destructive or paid actions; existing explicit session authorization still applies.

Mode

Plan a controlled comparison; apply for requested experiment code or bounded execution.

Prerequisites

Dataset/split manifests, fixed objective/metric, environment/dependencies, baseline where applicable, and explicit compute limits. Record code revision, configuration, seeds, artifact paths, and resource use. Local smoke checks do not imply authorization for paid training. Never optimize on the held-out test set.

Expected output

• Ablation protocol or run results as a variant matrix with attributable differences, uncertainty, cost/quality interpretation and caveats.

How the work is checked

• Data changes do not confound the component comparison; noisy score differences are not declared decisive.

When to stop or clarify

• Do not remove interdependent components without explaining the changed model contract. Budget limits constrain repeats and confidence.

Handling missing context

Infer

Read framework, training entry point, loss/metric, split manifests and checkpoint conventions from supplied source.

Assume

In apply mode, implement requested code and tiny isolated smoke checks with existing tools, and otherwise propose them; leave unmeasured model quality explicit.

Ask

Ask for unresolved objective/data semantics before encoding them, and environment/resource limits before launching training or a search; implementation alone does not need a hardware purchase decision.

Technical guidance

Evidence

Identify the component claim, matched data/protocol, randomness and comparison metric.

Method

Remove or vary one meaningful component while holding the rest fixed and repeat enough to expose relevant variance within budget.

Pitfall

A changed preprocessing pipeline or compute budget can confound a claimed component improvement.

Check

Compare paired outcomes where possible, report uncertainty and preserve a no-change control when evaluation noise matters.

Situational decisions

When removing a component changes another required contract: Redesign the comparison or disclose the confound instead of attributing all change to one component.

When the request is for local preparation or implementation: Implement isolated feature/configuration switches and comparable run metadata; defer result claims until comparable measurements exist.

The coding agent follows this workflow using its available tools. Installation does not grant service access or guarantee an outcome. Read the compatibility notes.

Keep exploring