lllllllama/rigorpilot-skills
11 skills2M installs5.6k starsGitHub
Install
npx skills add https://github.com/lllllllama/rigorpilot-skillsSkills in this repo
1Ai Research ExploreA research exploration skill for deep learning candidates within a durable research context. Use it when you have a defined task, dataset, and evaluation method and want to systematically explore ideas while maintaining scientific rigor and comparability.176kinstalls2Analyze ProjectA read-only analysis skill for deep learning research repositories. Use it when you need to understand model structure, training flow, and configuration before making changes. It analyzes code without modifying it and identifies likely insertion points for improvements.176kinstalls3Ai Research ReproductionA repository reproduction skill for deep learning projects. Use it when you want a minimal trustworthy run that reads the README first, selects the smallest documented target, and records all evidence and deviations.176kinstalls4Explore CodeA code-level exploration skill for deep learning research. Use it when you want to make targeted code changes like adding adapters, replacing heads, or transplanting modules while maintaining isolation and audit trails.176kinstalls5Paper Context Resolverpaper-context-resolver is the Rigor Paper Context helper in rigorpilot-skills for README-first deep learning reproduction. It activates only when README and repository files leave a narrow reproduction-critical gap, then resolves specific paper details such as dataset splits, preprocessing steps, evaluation protocols, checkpoint mapping, or runtime assumptions from primary sources while recording conflicts. Developers reach for paper-context-resolver during ai-research-reproduction when documented repo guidance is incomplete but the missing detail is essential to a trustworthy run. The skill is not for general paper summaries, title-only lookups, environment setup, command execution, or replacing README guidance by default.176kinstalls6Safe DebugA debugging skill for deep learning research. Use it when you have a traceback, training failure, or error symptom and want conservative diagnosis before any code changes.176kinstalls7Repo Intake And Planrepo-intake-and-plan is the Rigor Intake helper in rigorpilot-skills that scans a repository, reads README and common project files, extracts documented commands, and classifies inference, evaluation, and training paths. It checks nine primary files when present—including README.md, requirements.txt, pyproject.toml, setup.py, and Dockerfile—and inspects high-signal directories such as configs, scripts, and experiments for command clues. Developers reach for repo-intake-and-plan when onboarding to an unfamiliar deep learning codebase and deciding the smallest trustworthy reproduction or modification target before setup or execution. The output is a conservative recommendation rather than code changes, making it the usual first step before env-and-assets-bootstrap or ai-research-reproduction.176kinstalls8Run TrainA training execution skill for deep learning research. Use it when you have a selected training command and want conservative execution with structured status, checkpoint, and metric reporting.176kinstalls9Explore RunAn exploratory execution skill for deep learning research. Use it when you have explicit authorization for exploration and want to run small-subset validation, sweeps, or quick trials with fair-comparison caveats.176kinstalls10Minimal Run And Auditminimal-run-and-audit is the Rigor Run leaf mode from rigorpilot-skills for developers validating agent inference and evaluation pipelines before trusting results. The skill runs a selected smoke, inference, evaluation, or sanity command conservatively, captures execution evidence, and writes four artifacts: SUMMARY.md, COMMANDS.md, LOG.md, and status.json. Reporting policy requires separating facts from inferences, naming the documented command explicitly, and stating whether the run was full, partial, smoke-only, sanity-only, or blocked. Developers reach for minimal-run-and-audit when an agent change needs a reproducible audit trail instead of ad-hoc terminal output, especially after patches or when a blocker must be surfaced clearly in SUMMARY.md without burying the root cause.176kinstalls11Env And Assets Bootstrapenv-and-assets-bootstrap is the Rigor Setup mode in rigorpilot-skills for conservative environment and asset preparation before a reproduction run. The agent follows a four-level evidence order—README links and paths, config files and default arguments, code-level constants, then careful filename inference—to prepare checkpoints, datasets, and caches transparently. It uses a conda-first environment strategy and records source URLs or identifiers when downloading assets while avoiding mismatched assumptions. Developers reach for this skill after repo-intake-and-plan selects a target and before run-train or ai-research-reproduction execution steps. The goal is reproducible dependency and asset readiness grounded in repository documentation rather than speculative package installs.176kinstalls