Second Brain

kepano/obsidian-skills evaluation — defuddle is the only upgrade candidate

2 min read

kepano/obsidian-skills evaluation — defuddle is the only upgrade candidate

Decision

From the 5 skills in kepano/obsidian-skills, only defuddle (web page → clean markdown via Mozilla Readability) adds value. The other 4 (json-canvas, obsidian-bases, obsidian-cli, obsidian-markdown) are Obsidian-specific — no Obsidian in our stack, Excalidraw/Vault MCP/Graphiti cover their domains. Pending: install defuddle (npm install -g defuddle) and create a lightweight skill guiding when to use defuddle parse <url> --md vs WebFetch. Architecture.md drift (phantom obsidian skill references) already fixed.

Rationale

4 of 5 skills require Obsidian as renderer/runtime — not in our stack, would add dead capability. defuddle is a standalone CLI (Mozilla Readability algorithm) that strips nav/ads/clutter from web pages → clean markdown. Reduces token usage on web content extraction. Aligns with Law 7 (Efficiency) and feedback_cost_efficient_execution (deterministic tool > LLM). Zero infrastructure cost (npm package).

Alternatives Rejected

  • Install all 5 skills: Rejected — 4 would be dead code with no renderer (Law 3: Zero Technical Debt)\n- Install none: Rejected — defuddle genuinely enhances token efficiency for web research\n- Build custom web extraction: Rejected — defuddle already implements Mozilla Readability, no need to reinvent

Outcome

SUCCESS (verified 2026-04-16). defuddle installed globally at /usr/bin/defuddle via npm. The 4 Obsidian-specific skills were correctly not installed — no Obsidian runtime in stack, zero dead code. Remaining: lightweight skill/guide for when to use defuddle parse <url> --md vs WebFetch not yet created (low priority, defuddle works standalone via CLI).

Graph View

Backlinks