Menu
SkillsMP has collected 30 skills from Tyler-R-Kendrick/slm-training. Open a skill to review its source and details.
Showing 30 of 30 collected skills.
Discover and run the reasoning/revmath profile — reverse-mathematics / computability analysis over existing campaign, formal-preflight, evidence, and repair owners. Use when editing revmath schemas, fixtures, labeling, self-healing, four-axis ledgers, or…
Use when the Linear In Review queue for team SLM is backed up and an agent session should claim the oldest issues, finish each in an isolated git worktree (reusing open PRs), fix CI in bounded rounds, and squash-merge once merge preflight passes
Use when a scheduled or interactive session should report the current autotrain evidence state — open hypothesis families, closed approaches, preflight blocks, pending confirmations, promoted-model status — as a short read-only markdown brief in chat
Use when refreshing the generated OpenWiki pages under docs/openwiki — a scheduled or manual session installs the pinned OpenWiki CLI, selects a provider key, runs scripts.update_openwiki, and opens a PR on branch openwiki/update (skips cleanly when no…
Use when a screening positive appears in the continuous autotrain loop and before any confirmation spend — convene three independent skeptic lenses (statistical power, prior evidence, mechanism plausibility) to attempt refutation; majority-refute files the…
Operate the OpenUI SLM training pipeline end to end, including a continuous hands-off model and harness improvement loop. Bare /autotrain is non-terminating and must not stop for user confirmation; an explicit phase or --once is finite. Code fixes during…
Use when evaluating models, writing or interpreting ship gates, claiming readiness, changing parse/fidelity/reward metrics, or deciding fixture-demo vs production ship
How to do multi-step engineering work in this repo: parent agents plan and stack PRs with official GitHub Stacked PRs (`gh stack` / `gs`); subagents implement layers with incremental check-ins; parent closes every PR bottom-up with rubber-duck adversarial…
Improve, extend, debug, or review any canonical OpenUI training harness, including autoresearch/hypothesizer, annotations, distillation, experiment selection/promotion, model build/evaluation, preference learning, quality/retrieval, RL, held-out test data, or…
Run or extend evidence-grounded OpenUI autoresearch and autotraining campaigns, including literature discovery, typed experiments, data repair, telemetry, researcher evaluation, persistence, and RL readiness.
Diagnose and improve Lean4-calculated metric ranges when an autotraining, evaluation, benchmark, or promotion observation is outside its preregistered band. Use for metric_evidence/v2 and metric_certificate/v2 work, LeverProof theorem or assumption changes,…
Use when running, extending, or interpreting quality (E*), grammar (X*), perf (P/Q/R), phase, scaling, or mixture experiment matrices
Run the knowledge-driven OpenUI research loop that coordinates the training pipeline with curated knowledge. Read and update the repo + personal brains (OKF / Obsidian) and OpenWiki, run the prior-work discovery loop to find related research, drive the…
Use when finishing any train, eval, benchmark, profile, telemetry, matrix, or reproduction run — or when a checkpoint was created/promoted without MODEL_CARD/README updates — or when results exist only under outputs/ or in chat without matching docs/design…
Close the data-quality loop after any training-data synthesis or build. Use whenever build_train_data (or the build_train_data job) runs, when a quality_report.json shows warnings, when planning changes to producers/synthesizers, or when deciding what data…
Use whenever a dashboard page changes — editing any src/apps/dashboard/src/pages/*.tsx or the shared components in src/apps/dashboard/src/components.tsx, or when adding/removing a route in main.tsx. The dashboard renders every page two ways (a…
Create or refresh committed train-only frontier description bundles for OpenUI gold records. Use when filling src/slm_training/resources/frontier/worklist.jsonl with paraphrases, L1-L5 abstraction prompts, minimal edit instructions, or optional external-page…
Keep this repository's tracked files and directories canonical, navigable, and free of redundant copies. Use before creating, moving, renaming, deleting, or duplicating tracked files or folders; when adding modules, scripts, docs, fixtures, tools, skills,…
Autonomous NeMo-RL research agent workflow for directed hypothesis testing and open-ended discovery. Guides agents through the full experiment lifecycle: understanding recipes and environments, wiring RL or NeMo-gym runs, launching reproducible baselines and…
Safely plan, submit, and reconcile pinned NVIDIA NeMo RL training on Hugging Face Jobs with durable OpenUI artifacts. Use for NeMo RL Jobs commands, recipes, credentials, checkpoint handling, or hardware-smoke claims.
Manage durable working-session memory for coding agents. Use when a user asks to preserve or recover agent context across disconnects, VS Code restarts, long-running work, handoffs, or any session where important state should be written periodically under the…
Token-reduction skill for coding agents. Compresses JSON tool output, build logs, grep/ripgrep search results, source code, and conversation history the same way the Headroom proxy does — but in-context, with no binary install. Use this skill whenever an…
Hugging Face Hub CLI (`hf`) for downloading, uploading, and managing models, datasets, spaces, buckets, repos, papers, jobs, and more on the Hugging Face Hub. Use when: handling authentication; managing local cache; managing Hugging Face Buckets; running or…
Plan and coordinate the deployment of a model to Amazon SageMaker AI. Use this skill whenever the user wants to deploy, host, serve, or expose a model on SageMaker or AWS — including phrases like "deploy a model", "host this LLM on AWS", "serve this embedding…
Create a SageMaker endpoint (real-time or async) with autoscaling, CloudWatch alarms, and tagging enabled by default. Use this skill whenever about to create a SageMaker endpoint, write deployment code that calls `create_endpoint`, or finalize a deployment…
Use when the user asks about finding the best, top, or recommended model for a task, wants to know what AI model to use, or wants to compare models by benchmark scores. Triggers on: "best model for X", "what model should I use for", "top models for [task]",…
Build and publish a Gradio demo on Hugging Face Spaces for a user-provided LoRA. Use when someone asks to create, generate, ship, or publish a Space, demo, Gradio app, or playground for a LoRA — including LoRAs for Qwen-Image, Qwen-Image-Edit, LTX-Video, Wan,…
Build, deploy, and maintain applications on Hugging Face Spaces — Gradio / Docker / Static SDKs, ZeroGPU and dedicated hardware, model loading, debugging, buckets, inference providers, community grants. Use whenever the user asks to create or host an app on…
Rust Token Killer (rtk-ai/rtk) — compress shell/tool command output by 60–90% before it enters the LLM context. Use for git status/diff/log, pytest, ruff, ls, docker, kubectl, and other verbose CLI output. Prefer `rtk <cmd>` over raw commands when rtk is…
Train and fine-tune transformer language models using TRL (Transformers Reinforcement Learning). Supports SFT, DPO, GRPO, KTO, RLOO and Reward Model training via CLI commands.