ai-research-explore Skill: Governed ML Research Experiments

ai-research-explore makes your agent run candidate-only deep learning experiments on a research repo, with idea gating and fair comparison.

Installs: 345.4K; on skills.sh
Stars: 486; on GitHub
Files: 33; besides SKILL.md
Updated: Sep 2026; last repo commit

Use it only when you explicitly authorize exploration and have a durable current_research anchor, such as a branch, commit, checkpoint or run record. The agent freezes the task, dataset, benchmark, evaluation source, SOTA reference and compute budget, maps your ideas to sources, ranks them with explicit gates, then tries one bounded candidate at a time.

Output goes to analysis_outputs/, sources/ and explore_outputs/, including SCIENTIFIC_CHANGELOG.md and COMPARABILITY_REPORT.md. SKILL.md says novelty stays a hypothesis and the skill does not promise autonomous discovery. It hands narrow code or run work to explore-code and explore-run, so install the whole set.

Good for

Researchers who have picked a task, dataset and benchmark and want ranked candidate ideas
ML engineers trying bounded variants on top of a working training branch
Teams that need exploratory gains labeled as candidates, with comparability notes

Install ai-research-explore

Claude Code

Copy this prompt.

Install the ai-research-explore skill in Claude Code. Run this command:

npx skills add lllllllama/RigorPilot-Skills --all

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

This skill reads shared guideline files from the ai-research-reproduction skill, so install the whole set rather than this folder alone. The README says the main reproduction skill works alone and that "all skills" is the choice for the companion and leaf skills.

From the project README. Source.

What's in the skill

When it loads

Rigor Explore compatible skill slug for meaningful and potentially novel deep learning research candidates. Use when the researcher has chosen the task family, dataset, benchmark, evaluation method, provided SOTA references, and wants candidate-only exploration on top of `current_research` with auditable repo understanding, idea gating, fair comparison, and governed experiments written to `explore_outputs/`. Do not use for README-first trusted reproduction, open-ended direction finding, narrow code-only or run-only exploration, passive repo analysis, verified novelty claims, or implicit experimentation.

Sections in SKILL.md

Purpose
Fit
Research Rhythm
Workflow
Ranking and Evidence
Campaign Inputs
Reference Loading

Key facts

Author: lllllllama (owner of the lllllllama/rigorpilot-skills repo)
License: MIT
Installs on skills.sh: 345,396 as of 2026-11-02
GitHub stars: 486 for the whole lllllllama/rigorpilot-skills repo
skills.sh audits: Gen Agent Trust Hub warn, Socket warn, Snyk warn
What is inside: 33 files besides SKILL.md (26 in scripts/, 6 in references/, 1 in agents/)
Needs: The other RigorPilot skills (explore-code, explore-run, analyze-project, run-train) and shared files from ai-research-reproduction; a research anchor; Python scripts run on your machine
Last commit to the repo: 2026-09-23

What it includes

Stops if either is missing
Needs explicit exploration permission and a current_research anchor. SKILL.md says to use this skill only when both are present.
Read · Confirm the anchor and your authorization
Task, dataset, benchmark, evaluation source, SOTA, budget
Takes a research_campaign or the older variant_spec. Optional fields such as candidate_ideas and idea_policy guide it but are not required.
Try: “Use ai-research-explore on my current branch: dataset X, benchmark Y, compute budget 8 GPU-hours, try bounded candidates.”
Read · Campaign inputs
Builds only the repo notes the campaign needs
Usually through analyze-project, writing to analysis_outputs/.
Files · Repo understanding
Cache-first lookup of papers and sources
Prefers local curated literature such as Zotero, then seed sources, repo-local and public locators, then optional web lookup. It treats lookup as source resolution, not open-ended search.
Network · Bounded source lookup
Keeps your ideas, scores them with explicit gates
Ranks before running by expected gain, cost, success likelihood, patch surface, dependency drag, evaluation risk and rollback ease. It stops for a checkpoint when top ideas are too close.
Write · Idea gating and ranking
Hands code work to explore-code, runs to explore-run
Uses minimal-run-and-audit or run-train only when the plan needs real execution evidence, then re-ranks by observed results.
Runs code · One candidate at a time
Deterministic workflow and artifact writer
With scripts/write_outputs.py it writes candidate-only outputs and the changelog and comparability reports.
Runs code · scripts/orchestrate_explore.py

From the official docs

Permissions

Medium risk

The agent edits code and runs experiments on your machine with no sandbox. Keep it on an isolated branch and set a compute budget first.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It edits code and runs experiments for you
Exploration requires your explicit authorization and should stay on an isolated branch. The README says approved commands can access the host and network and to use trusted repositories. Check the branch before any run starts.
Experiments cost compute time
A campaign has a compute budget and caps such as max_variants. Set both before you start.
Results are candidates, not proof
SKILL.md says novelty and gains stay hypotheses until literature contrast, ablations and fair comparison.

Technical details

Talks to	Source lookup (local library first, public locators, optional web lookup)
Scopes	Runs bundled Python scripts and training or evaluation commands in your repo Writes analysis_outputs/, sources/ and explore_outputs/ May edit code in an isolated branch or worktree

Questions

What does ai-research-explore do?

It runs governed, candidate-only experiments on a deep learning repo. It freezes your campaign inputs, gates and ranks ideas, tries bounded changes and writes outputs that mark every gain as a candidate, not a verified result.

How do I install it?

Use the install box at the top of this page, which installs the whole RigorPilot set. The skill reads shared files from ai-research-reproduction, so installing it alone is not enough. The README says the installer needs Node.js/npm and that skills@1.5.26 requires Node 22.20.0 or newer.

When should I not use it?

Not for README-first reproduction, open-ended direction finding, narrow code-only or run-only work, passive repo analysis, or any claim of verified novelty. SKILL.md sends those to ai-research-reproduction, explore-code, explore-run or analyze-project.

Does it need an API key?

No key is named in SKILL.md. It works inside your agent and its Python helpers run on your machine.

How is it different from explore-code and explore-run?

ai-research-explore coordinates a whole campaign on top of current_research. explore-code handles only exploratory code changes and explore-run only exploratory runs, and it uses those two for the work.

Is it safe?

It edits code and runs experiments on your machine, and the README says this is not an OS sandbox. SKILL.md keeps exploration isolated from the trusted baseline and needs your explicit authorization. See the audit results under key facts.

Information

updated Nov 2, 2026

Repository: lllllllama/rigorpilot-skills
Folder: skills/ai-research-explore

lllllllama/rigorpilot-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 3

RigorPilot Skills README · github.com
ai-research-explore SKILL.md · github.com
skills.sh listing and audits · skills.sh