Comparison: Obsidian Skills vs Agentic Memex

Short answer

kepano/obsidian-skills is a high-quality tool-use layer for Obsidian syntax and operations. Agentic Memex should be a higher-level intelligence architecture: entity schema, provenance, retrieval, benchmarks, and operating workflows.

Positioning

Dimension	Obsidian Skills	Agentic Memex
Primary job	Help agents create/edit Obsidian files correctly	Help agents reason over a citeable knowledge graph
Scope	Markdown, Bases, Canvas, Obsidian CLI, web extraction	Entity model, ingestion, curation, retrieval, briefing, evals
Unit of work	File syntax and vault operations	Claims, entities, relationships, decisions, sources
Data model	Obsidian flavored markdown conventions	Typed entities with provenance and lifecycle
Benchmarking	Not central	Core requirement
Privacy posture	General-purpose vault tooling	Explicit sanitized/public framework, private data excluded
Best use	Editor ergonomics	Strategic/analytical memory system

What to borrow

• Agent Skills compatibility.
• Small, focused SKILL.md files.
• Clear command recipes.
• Obsidian-native markdown discipline.
• Separate skills for different operations.

What to avoid

• Staying at the syntax-helper layer.
• Treating the vault as just notes, not as a knowledge substrate.
• Shipping without evals.
• Shipping examples that leak private relationships or real strategy.

Recommended architecture

Agentic Memex should sit one level above Obsidian Skills:

Obsidian / Markdown editor layer
        ↑
Obsidian Skills: syntax, bases, canvas, CLI
        ↑
Agentic Memex: entity schema, provenance, graph, retrieval, evals, briefings
        ↑
User-facing agent: answers, decisions, prep, memory maintenance

Interoperate (use them together)

MemexLab and kepano/obsidian-skills (MIT, by Steph Ango / kepano) both follow the Agent Skills specification, so they install side by side in the same agent and compose cleanly — obsidian-skills handle the vault / file layer, MemexLab handles the knowledge layer.

Install both (works for OpenClaw / Claude Code / Codex / OpenCode agents):

npx skills add kepano/obsidian-skills    # obsidian-markdown, obsidian-bases, json-canvas, obsidian-cli, defuddle
# then add MemexLab's skills/ to the same agent skills directory

How they compose:

kepano/obsidian-skills	…feeds / supports	MemexLab
defuddle (clean HTML → markdown)	source capture	memex-ingest / memex-extract
obsidian-markdown	correct file syntax for the items MemexLab writes	memex-markdown
obsidian-bases	tabular views over MemexLab's typed frontmatter	schema / memex-progress coverage
json-canvas	visual maps of the knowledge graph	concept maps / relationships
obsidian-cli	vault file operations	all write paths

Use obsidian-skills for operating Obsidian correctly; use MemexLab for operating the knowledge. We link to obsidian-skills rather than bundling it — see its repo for current skills and install.

Benchmark axes

1. Retrieval precision: did the system find the right entities and sources?
2. Citation quality: are claims tied to actual source notes?
3. Synthesis quality: does the answer sharpen the user’s decision?
4. Deduplication: does it merge duplicate people/companies/sources safely?
5. Contradiction handling: does it flag conflicts instead of smoothing them over?
6. Privacy safety: does it avoid leaking private material into public outputs?