Skills lllllllama rigorpilot-skills
RigorPilot Bundle of 11 skills
- Use it when:
- You want an agent to run a deep learning research repo from its README and record honestly what worked.
- It does:
- Finds the smallest documented run, such as one evaluation command, sets it up, runs it and audits the result, instead of a full training run.
- You get:
- A repro_outputs/ folder: a verdict (verified, partial or blocked), the exact commands, logs, patches, any change to data, evaluation or metrics, and the README annotated section by section.
- Main skill:
- ai-research-reproduction runs a reproduction end to end and works alone; six of the others are steps it uses as needed, and each can be called by name.
- Exploration lane:
- ai-research-explore, explore-code and explore-run try new ideas on a saved baseline, only when you ask, and never report them as reproduced.
- Also has:
- safe-debug, which diagnoses a failed run and patches only with your approval.
- Needs:
- Python 3.11+ and Git for the main skill's scripts; five of the other skills load a runtime, writer or reference from it, so install it with them.
TL;DR
ai-research-reproduction is the entry point, and the README offers it as a self-contained install. It picks the smallest documented target (inference, then evaluation, then a training startup check; full training only after you confirm) and runs it through its script in three calls: plan, run, verify. Six skills are steps it uses as needed, and each can also be called by name. repo-intake-and-plan sorts the README's commands and recommends the first run. env-and-assets-bootstrap plans a conda environment and where checkpoints and datasets go. paper-context-resolver answers one gap the README leaves, such as a dataset split, from the paper, and notes any conflict. analyze-project maps a repo's entry points and configs, read-only, when its structure is unclear. minimal-run-and-audit records an inference or evaluation run in repro_outputs/, and run-train starts, checks or resumes training, recording it in train_outputs/. When a run fails, safe-debug diagnoses it and patches only with your approval. Exploration is a separate lane, only when you ask and from a saved baseline: ai-research-explore ranks ideas and hands code changes to explore-code, which works on an isolated branch, and short trials to explore-run. Their results go to explore_outputs/ as candidates, never as reproduced results.
Summarizes the repo at fb3ccdf.
Any agent, with the skills CLI
npx skills add https://github.com/lllllllama/rigorpilot-skills --skill '*' Source on GitHubAll skills from lllllllama/rigorpilot-skills
- Total Downloads
- 492,779 of paper-context-resolver, its most installed skill, on skills.sh, Nov 9, 2026
- Repo stars
- 486 lllllllama/rigorpilot-skills
- First seen
- Mar 30, 2026 on skills.sh, its first skill
Its skills, in the order they're used
The ones marked “Only in the bundle” need the rest of it. The others also work alone.
-
ai-research-reproduction
353K total downloadsReproduces a deep-learning repo from its README through the smallest documented run, and records the evidence.
-
repo-intake-and-plan
492K total downloadsMaps a deep-learning repo and plans a README-first reproduction.
-
paper-context-resolver
493K total downloadsFills the gaps a README leaves when reproducing a deep-learning paper's repo.
-
env-and-assets-bootstrap
492K total downloadsPrepares a conda environment, datasets and checkpoints for a reproduction.
-
analyze-project
353K total downloadsMaps a deep-learning research repo's entry points, configs and suspicious code patterns, read-only.
-
Runs a minimal reproduction and records the evidence.
Its run and report scripts load a shared runtime that ships inside ai-research-reproduction, the repo's main skill, and stop with an error without it.
-
Runs a chosen training command conservatively and records its logs, checkpoints and metrics.
Its run and report scripts load a shared runtime that ships inside ai-research-reproduction, the repo's main skill, and stop with an error without it.
-
safe-debug
352K total downloadsDiagnoses deep-learning training and inference failures and proposes the smallest fix, patching only with your approval.
-
Explores novel research candidates on top of a reproduced baseline.
Loads its required research loop and shared runtime from ai-research-reproduction, the repo's main skill, and hands code changes and trial runs to the repo's explore-code and explore-run skills.
-
Makes candidate research code changes, such as LoRA layers or a new head, on an isolated branch with rollback notes.
Its report script loads a shared writer that ships inside ai-research-reproduction, the repo's main skill, and stops with an error without it.
-
Plans and ranks small exploratory runs, such as subset checks and short sweeps, on top of a research baseline.
Its report script loads a shared writer that ships inside ai-research-reproduction, the repo's main skill, and stops with an error without it.