Skip to content
Get packs

Ablation Planner

Plans reviewer-proof ablation studies for a paper, then helps run them

Source Download skill
Repository
wanshuiyin/Auto-claude-code-research-in-sleep
Stars
18,482
Licence
MIT
Files
1 file

What it does

When an AI agent has installed this skill, it can plan the ablation experiments a research paper needs. An ablation study removes or swaps out one part of a method to show whether that part really matters. The skill asks a Codex model (via an MCP connector) to act as a strict reviewer and propose a prioritized list of ablations, hyperparameter sweeps, and design comparisons, with expected outcomes and a run order. The agent then checks compute budget and code changes, creates configs or scripts for each ablation, and runs them while logging results.

Great for

  • ML researchers preparing a paper submission whose main results already hold up
  • Research engineers who want a checklist of experiments reviewers are likely to ask for
  • Users of the wider Auto-claude-code-research-in-sleep workflow who already use its result-to-claim and review-loop skills

Skip it if

  • Projects without a validated main result: the skill is meant to run after results pass a claim check, so it adds little earlier on.
  • Anyone expecting it to run standalone: it assumes project files such as research contract, experiment logs and a Codex MCP server are present, and it can launch real GPU jobs that cost money and time.
  • Non-research code: the workflow is written around paper reviewers and ML experiments.

Try asking

  • “Main results passed the claim check; plan the ablations a reviewer would ask for on my method.”
  • “Given this method description, list the must-run component ablations and their expected outcomes.”
  • “Estimate the GPU-hours for the planned ablations and propose cuts if the budget is tight.”
  • “Create configs for the config-only ablations and log the results in EXPERIMENT_LOG.md.”
Step-by-step guide: how to use Ablation PlannerInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the MIT licence and an attribution note.

    ablation-planner.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip ablation-planner.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip ablation-planner.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Run the claim-validation step first so the skill only starts once main results are supported or partially supported.
  • Keep the Codex model name and reasoning settings in the skill up to date with the models your account actually has access to.
  • Review the generated run order and compute estimate before letting the agent launch any jobs.

What's inside

  • SKILL.md

Redistributed unmodified from wanshuiyin/Auto-claude-code-research-in-sleep under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.