Claude Skills by VectorSpaceLab

Analyze the gap between a recovered paper result and the paper target, decide whether generated skills are acceptable, and produce actionable module-level feedback for another distillation cycle.

Convert one paper module document into a production-quality English Agent Skill with scripts, tests, validation logs, and clear input/output contracts.

Creates reusable Agent Skills from an AI research paper using the Paper2Skills Distiller workflow. Use in Creator mode when the user provides a paper PDF, paper URL, arXiv id, paper title, or paper/repo pair, or asks to convert a scientific paper into skills with recovery experiments. Prefer a TOML run config; ask for missing paper, repo, recovery, runtime, and budget fields before expensive work.

Creates a repo-specific operating Agent Skill for DisCo Researcher from a local repository by inspecting source files and an installed or auto-prepared, backend-aware Python package environment. Use when the user asks to create a skill for a repo, generate repo-specific skills, analyze a local Python package, or build a DisCo skill that may also be exported to compatible agents. If the user gives only a repo path, analyze the repository structure and native tests/examples first, confirm the e...

Designs and validates a reusable Creator meta-skill workflow for an evidence-backed recurring construction gap after distill-ml-knowledge selects the design-reusable strategy. Use it to parameterize future source or task anchors, verify the workflow, and prepare an approved managed Creator-infrastructure import.

Canonical Creator entry point for DisCo skill distillation. Use when turning a repository, paper, tutorial, dataset, benchmark, research note, task, or other source material into a verified operating skill graph. Identify the anchor, scope capabilities, ground them in evidence, construct a candidate graph, verify it into an accepted graph, and record the construction.

Extends an existing repository-specific Agent Skill with new capabilities, deeper coverage, troubleshooting, scripts, and usability tests. Use when the user asks to expand, improve, deepen, or add coverage to an already implemented skill instead of creating a new skill from scratch. If the repository itself changed and the old skill may be stale, use refresh-repo-skill instead.

Export all or selected DisCo-managed repository skills, their regenerated repository-index.jsonl, and a scoped area/family repo-skills-router into Codex, Claude Code, or another agent skills directory. Use when the user asks to copy, install, synchronize, or merge repository skills into another agent or project. Resolve exact skill IDs, obtain explicit approval for replacements, then use the bundled transactional helper for validation, Codex policy injection, deterministic router merging, rol...

Orchestrate a paper-to-skills distillation loop for AI research papers supplied as local files, URLs, arXiv ids, or titles, optionally with a local or remote repository, by resolving sources, modularizing the paper, creating module skills, preparing recovery runtime, recovering a paper result without the source repo, analyzing gaps, and iterating until acceptable.

Analyze a local AI research paper, optionally using a local repository as supporting evidence, and produce a concise paper profile plus a validated module plan of at most five reusable skill modules.

Inspect, prepare, and document an isolated runtime environment for Paper2Skills recovery experiments, including package probes, model/cache checks, GPU status, bounded benchmark acquisition, and a runtime handoff for recover-paper-result.

Create, install, and prove the minimum backend-aware Python inspection environment set for a local repository before create-repo-skill. Use when a repo package must be installed into new or existing Conda, micromamba, uv, or venv environments; when selected native tests/examples require CPU, CUDA, ROCm, MPS, or vendor accelerator dependencies; when a broken inspection environment needs diagnosis; or when hardware/backend compatibility must be verified. Consume the confirmed extraction scope a...

Reproduce a fast paper result using only the paper, module documents, generated module skills, and datasets, explicitly forbidding reads from the original source repository during recovery.

Refreshes an existing repository-specific Agent Skill after the source repository changed. Use when the user says repo code, APIs, docs, examples, configs, dependencies, or behavior changed and an old skill may now be stale, outdated, inconsistent with current code, or needs to be resynchronized from repository evidence.

Routes substantive ML, AI, data, scientific-computing, and software-engineering requests to the smallest useful set of managed repository skills. Invoke proactively when a request names or implies a package, framework, model family, dataset, modality, workflow, backend, deployment target, evaluation method, or implementation approach that may benefit from repository guidance, even if no repository is named. Narrow progressively from area to family to repository root: inspect only the one or t...

Verifies a generated or refreshed repo-specific Agent Skill by creating assertion-backed usability test cases, running content-level self-refine, checking backend-classified native repo examples/tests against the prepared CPU/GPU environment plan, enforcing required-backend and import gates, checking static quality, and producing final coverage and handoff artifacts. Use this after create-repo-skill, refresh-repo-skill, or extend-repo-skill finishes an integrated runtime skill draft, and when...

Writes deterministic DisCo workflow scripts that coordinate subagents with stable coverage IDs, bounded recovery, explicit usage semantics, and structured environment handoffs.

Use the Hugging Face Hub Python client and hf CLI for repository and artifact management, downloads and caching, hosted inference, cloud compute, model integration, and safe automation.

Routes reliable Hugging Face Hub CLI and shell automation, including hf output modes, authentication, file movement, repository and catalog operations, hosted jobs, extensions, and agent-skill generation. Use for hf command selection, JSON or quiet parsing, login and token handling, dry-run or confirmation planning, extensions, or compatible CLI skills.

Routes Hugging Face Hub file and snapshot downloads, cache and offline recovery, Xet-backed storage, HfFileSystem and hf:// URIs, buckets, and safe copy or sync planning. Use for model, dataset, Space, or kernel downloads; allow/ignore patterns; cache corruption; fsspec browsing; volume mounts; and bucket file movement.

Routes hosted Jobs and Sandboxes, Space runtime configuration, OAuth and webhooks, framework/model integration, repository cards, tensor serialization, and TensorBoard logging workflows.

Operate Hugging Face Hub repositories and API resources with authenticated or mocked Python workflows, explicit revision control, safe mutations, and actionable error recovery.

Run hosted model inference and manage Hugging Face Inference Endpoints with sync or async clients, provider-aware payloads, streaming, tools, structured outputs, MCP, and safe lifecycle recovery.

Routes substantive ML, AI, data, scientific-computing, and software-engineering requests to the smallest useful set of managed repository skills. Invoke proactively when a request names or implies a package, framework, model family, dataset, modality, workflow, backend, deployment target, evaluation method, or implementation approach that may benefit from repository guidance, even if no repository is named. Narrow progressively from area to family to repository root: inspect only the one or t...

"Routes 3D ResNets PyTorch video action-recognition workflows

"Prepare raw videos and annotations for the 3D-ResNets-PyTorch

"Routes training, fine-tuning, validation, checkpoint, and

"Routes 3DDFA_V2 face-alignment setup, still-image demos, video

"Use 3DDFA_V2 ONNX Runtime acceleration and CPU latency or speed

"Prepare 3DDFA_V2 runtime assets and native extension builds before

"Run 3DDFA_V2 still-image alignment, rendering, pose, texture, and

Run 3DDFA_V2 video, smoothing, and manual webcam tracking workflows.

"Guide 3DDFA Python inference, geometry rendering,

"Operate the optional 3DDFA C++ OpenCV DNN port, checkpoint-to-ONNX

"Reconstruct 3DMM vertices, serialize mesh outputs, and manage

"Operate 3DDFA Python image and video inference for landmarks,

"Routes 3DDFA training recipes, loss selection, checkpoint resume,

"Operate AB3DMOT 3D multi-object tracking workflows for KITTI and

"Routes KITTI and nuScenes data-layout, detection-conversion, and

"Evaluate, threshold, combine, and visualize AB3DMOT KITTI and

"Run AB3DMOT tracking safely and use the core AB3DMOT tracker APIs

"Use Hugging Face Accelerate for PyTorch training-loop migration,

"Use Accelerate big-model inference utilities for meta

"Save and resume Accelerate training state, register checkpoint

"Use this sub-skill when working with Hugging Face Accelerate

"Select, configure, and diagnose Accelerate distributed training

"Migrate raw PyTorch training and evaluation loops to Hugging Face

"Route Acme reinforcement-learning framework tasks across core

"Build and debug Acme core dm_env loops, specs, wrappers, logging,

Select, configure, and adapt Acme JAX agents and JAX experiment workflows.