I keep my Claude Code skills and Spawner knowledge skills here for an agentic Obsidian vault workflow.
.
├── claude-code/
│ ├── aristotle-prover/
│ ├── codex-code/
│ ├── convolutional-debate-agent/
│ ├── deep-research/
│ ├── geps-v5/
│ ├── llm/
│ ├── obsidian/
│ ├── ralph/
│ └── system-augmentor/
└── spawner/
├── arxiv-pdf-reader/
├── lecture-notes-sync/
├── mathpix-pdf-converter/
├── media-transcript/
├── obsidian-cli/
├── obsidian-notes-validator/
├── online-research/
└── ssrn-pdf-reader/
| Skill | What it does |
|---|---|
aristotle-prover |
Translates math problems into Lean 4 proofs through the Harmonic Aristotle API. |
codex-code |
Breaks work into subtasks, runs parallel Codex CLI agents in worktrees, then merges results. |
convolutional-debate-agent |
Runs adversarial multi-model debates and aggregates outputs with reliability weighting. |
deep-research |
Extracts content from YouTube, arXiv, SSRN, PDFs, and web articles, then synthesizes comprehensive research notes. |
geps-v5 |
Uses graph-guided evolutionary search and tournament ranking for portfolio strategy exploration. |
llm |
Routes prompts to any model across all providers — CLI-first, with auto-discovery and benchmark tracking. |
obsidian |
Automates vault CRUD, graph queries, and UI control with a 3-tier workflow. |
ralph |
Autonomous fresh-context loop for multi-model dissertation idea generation with saturation exit. |
system-augmentor |
Audits capability gaps, researches options, debates candidates, and applies the best upgrade. |
Spawner Skills (spawner/)
| Skill | What it does |
|---|---|
arxiv-pdf-reader |
Guidance for downloading and processing arXiv papers. |
lecture-notes-sync |
Patterns for syncing lecture recordings with structured notes. |
mathpix-pdf-converter |
Mathpix API patterns for PDF-to-Markdown conversion, including batch workflows. |
media-transcript |
Patterns for transcribing audio and video content. |
obsidian-cli |
Deep Obsidian CLI v1.12 reference with patterns, edge cases, and architecture decisions. |
obsidian-notes-validator |
Quality rules for validating notes in an Obsidian vault. |
ssrn-pdf-reader |
SSRN download patterns, including Cloudflare bypass gotchas. |
online-research |
Web research workflow patterns. |
I use two skill systems because they solve different problems. Claude Code skills are executable: each one is a slash-command skill with a SKILL.md workflow and optional scripts/, settings/, and references/ support files. Spawner skills are not executable; they are YAML plus Markdown knowledge packs that Claude reads as domain context. In practice, I combine them so a Claude Code skill does the actions while a Spawner skill supplies guardrails and known pitfalls.
Copy or symlink claude-code/<skill-name>/ into your Claude Code skills directory.
Run the skill by name from Claude Code (for example, /codex-code).
The skill's SKILL.md defines the execution workflow and any helper files it can use.
Load spawner/<skill-name>/ as a knowledge skill in your Spawner setup.
Use it as reference context during tasks; these files are guidance, not commands.
Pair a Spawner skill with a Claude Code skill when you want execution plus domain rules.
See the docs/ for detailed guides on each skill, the architecture, and getting started.
MIT