Loading...
Loading...
Found 16 Skills
Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, transferring GPT-4 capabilities to open-source models, or reducing inference costs. Covers temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.
Generate a portable, self-contained Agent Skill from mature, curated Obsidian wiki pages — turning a cluster of verified knowledge into a reusable "digital expert" (SKILL.md + references/). Use this skill when the user says "/vault-skill-factory", "make a skill from my wiki", "turn these pages into a skill", "generate an agent skill from my vault", "package my notes on X as a skill", "build a domain-expert skill from my wiki", or wants to distill recurring, mature wiki knowledge into a shareable skill. Inspired by OpenKB's "drop in a book → out comes a digital expert" pattern. The factory ONLY reads the vault and WRITES TO A REVIEW DIRECTORY — it never installs skills, never writes into .skills/, and never touches global skill directories.
The foundational knowledge distillation pattern for building and maintaining an AI-powered Obsidian wiki. Based on Andrej Karpathy's LLM Wiki architecture. Use this skill whenever the user wants to understand the wiki pattern, set up a new knowledge base, or needs guidance on the three-layer architecture (raw sources → wiki → schema). Also use when discussing knowledge management strategy, wiki structure decisions, or how to organize distilled knowledge. This is the "theory" skill — other skills handle specific operations (ingesting, querying, linting).
Bootstraps modular Agent Skills from any repository. Clones the source to `sources/`, extracts core documentation into categorized references under `skills/`, and registers the output in the workspace `AGENTS.md`.
Perform "Rank Reduction" on any domain — start from phenomena, extract dimensions, identify constraints, find irreducible independent generators (rank), and verify through generation tests and validation. Use when the user says "Rank Reduction", "find the rank", "what is the rank of this domain", or wants to find the irreducible principles of any domain.
Use when reducing model size, improving inference speed, or deploying to edge devices - covers quantization, pruning, knowledge distillation, ONNX export, and TensorRT optimizationUse when ", " mentioned.
Efficient AI techniques including model compression, quantization, pruning, knowledge distillation, and hardware-aware optimization for production systems.
Given a domain, identify the few independent forces that truly underpin it. Reduce dozens of phenomena to the minimal set of generators—only when you can regenerate all phenomena from these generators does it count. Use this when the user says 'rank reduction', 'find rank', 'what is rank', 'what supports this domain', 'what lies behind it', or wants to decompose any domain into its irreducible generators.
DISTILL
Human-led curation of accumulated metis and guardrails. Surface patterns across sessions, propose what to promote, compact, or dismiss. Use after multiple sessions, before a new phase, or when search results feel noisy.
Extract knowledge from closed tasks and archive context
Capture the current task into a structured temporary session bundle under `.agents/sessions/` so a learning agent can later distill durable repo knowledge. Use for completed, blocked, or abandoned tasks with meaningful changes, debugging, validation, or reusable lessons.