01
It starts by itself
you authorize exploration. Runs in Claude Code.
- ↻ you authorize exploration
explicit candidate-only request with a current_research anchor
lllllllama · 331K installsAI Research Explore skill
Runs by itself, you authorize exploration.
One ranked research candidate per run.
Install the AI Research Explore skill and make it a habit: follow https://gethabit.dev/skills/lllllllama/ai-research-explore.md
- 1Copy
- 2Paste into Claude Code, Codex or Cursor
- 3It runs by itself
331K installs484 starsMIT licenceSource on GitHub ↗
AI Research Explore
by lllllllama · 331K installs
The skill
The ai research explore skill runs candidate-only deep learning research on top of a current_research anchor, with repo understanding, idea gates, fair comparison and governed experiments in explore_outputs/.
Only when you ask.
The habit
One ranked research candidate per run.
↻ Runs by itselfyou authorize exploration
How the habit works.
You don't have to.
When does it start? How does it check its work? What does it remember?
One run, step by step.
Then it stops.
Stops after one candidate per inner loop, or when the compute_budget is used up.
- 01
Confirm the anchor
The agent confirms current_research and your explicit candidate-only authorization.
- 02
Freeze the campaign
It freezes task, dataset, benchmark, evaluation source, SOTA reference and compute budget.
- 03
Understand the repo
It builds only the repo artifacts it needs, usually with analyze-project.
- 04
Gate ideas
It keeps your ideas, may add a few single-variable ones and ranks them with explicit gates.
- 05
Run one candidate
It uses explore-code or explore-run, and minimal-run-and-audit or run-train only when real execution is needed.
- 06
Rank on evidence
It ranks by command status, metrics and smoke results, and stops for your checkpoint when top ideas are too close.
- 07
Write outputs
scripts/write_outputs.py writes candidate-only results, never as trusted reproduction.
What it may do.
Nothing more.
It writes only in its own folder and asks you before anything goes out.
- Writes only inanalysis_outputs/**, sources/**, explore_outputs/**, an isolated branch or worktree
- Talks tooptional web lookup for sources
- Sendsnothing
- Your OKIt asks you before anything goes out.
Is it for you?
Use it when it fits.
Read both lists. If the left one sounds like you, the AI Research Explore skill helps.
Use it when
- ✓ You test new model variants
- ✓ You run a bounded sweep
- ✓ You need auditable experiment records
Skip it when
- – You want a trusted reproduction
- – You have not chosen task and evaluation
AI Research Explore skill, answered.
Does it prove novelty?
No. Novelty stays a hypothesis until literature contrast, ablations and fair comparison support it.
When should I use another skill?
Use analyze-project for passive reading, explore-code or explore-run for narrow work, and ai-research-reproduction for README-first reproduction.
Will it pick between close ideas on its own?
No. If top ideas are too close, it stops for your checkpoint.
How do I install the AI Research Explore skill?
Copy the prompt at the top into your agent. It installs the skill from lllllllama/RigorPilot-Skills and sets up the habit. Or run npx skills add lllllllama/RigorPilot-Skills --skill ai-research-explore.
Is the AI Research Explore skill free?
Yes. It is published under MIT. Habit is free too.
AI Research Explore. Now as a habit.
Install the AI Research Explore skill and make it a habit: follow https://gethabit.dev/skills/lllllllama/ai-research-explore.md
All lllllllama skills →More planning skills →
Works in
- Claude Code
- Codex
- Cursor
- ChatGPT
- Claude app
Made by lllllllama, published under MIT. Habit links to the original and never changes it. Is this your skill?