Skip to content
Get packs

Result-to-Claim Gate

Judges whether experiment results support your claims and suggests the next step

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

What it does

After experiments finish, this skill has the agent gather results from sources such as W&B, experiment logs and a research contract, then send them to a secondary Codex reviewer agent. That reviewer judges which claims the numbers support, which they don't, and what evidence is still missing. The agent then routes to a next action: pivot, run supplementary experiments, or confirm the claim. It also runs a deterministic check that cited values exist in the files before the judgment.

Great for

  • Researchers deciding whether results justify claims before writing a paper or rebuttal
  • ML practitioners who want an independent second opinion on ambiguous experiment outcomes
  • Teams that run ablations and want a checkpoint before committing to conclusions

Skip it if

  • Sanity-check runs or early debugging, since the skill is designed for main results (the reviewer call adds cost and delay)
  • Situations needing a truly independent judgment: the file states the Codex review is same-family and marked provisional unless a different model family is used
  • Projects without experiment logs or a stated intended claim, because the reviewer has little to evaluate

Try asking

  • “Judge whether my latest ablation results support the claim that the new loss improves robustness”
  • “Check my W&B run results against the research contract and tell me what evidence is still missing”
  • “Decide whether I should pivot, add experiments, or confirm my main result before writing the paper”
Step-by-step guide: how to use Result-to-Claim GateInstall, 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.

    result-to-claim.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip result-to-claim.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip result-to-claim.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

  • Write the intended claim in idea-stage/docs/research_contract.md before running experiments so the judgment has a clear target
  • Keep EXPERIMENT_LOG.md and EXPERIMENT_TRACKER.md up to date, since the agent reads them as primary inputs
  • Treat a 'provisional' or same-family verdict as a useful signal to verify, not as final proof

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.