Skip to content
FSS Knowledge Base

Obsidian Automation Guide — Buy vs Piecemeal Decision

Section titled “Obsidian Automation Guide — Buy vs Piecemeal Decision”

Decision

Section titled “Decision”

Verdict: piecemeal upgrade, not the Obsidian Starter Kit (OSK, €199.99 Essentials / €249.99 Premium). Reached after 3 rounds of pushback from Talbot, each producing new verified evidence rather than re-assertion.

Reasoning that held up

Section titled “Reasoning that held up”
  • OSK’s AI-native layer (agents-with-identity, skills library, receptionist routing, AGENTS.md-as-SSOT) is architecturally the same pattern KB-OS already runs in production via dept JOB_DESCRIPTION.md/Staff/ charter chains + skills — not a gap to fill.
  • The transferable skill-writing methodology is published free (verified: “How I Build AI Skills That Actually Work” gives the actual 5-step recipe, not a teaser). What’s paid is ~375 skills written against Dubois’ PARA/Zettelkasten schema — schema-locked, doesn’t transfer to KB-OS’s dept-routed SMTM structure.
  • Factual correction found in round 3: “Hermes-like autonomy” (always-on, 24/7 agent) is not part of OSK at all — OpenClaw/Hermes Agent are separate, free, self-hosted projects Dubois runs alongside his vault. Buying OSK gets zero autonomy infrastructure.
  • Individual plugins Dubois sells alongside OSK (Ghost Publish, Typefully, Dataview Serializer, Kanban Action Planner) are independently free/MIT-licensed on his GitHub — only the OSK plugin itself (schema/note-type engine) is a private, paid-gated repo. → promoted as a general lesson (see below).

What came out of it (free, no purchase)

Section titled “What came out of it (free, no purchase)”
  • Related-Tasks.base / Related-Notes.base built at Core/Processes/Projects/KB-OS/Bases/ — native Obsidian Bases implementing the “No Plugins, No AI” clipping’s two-tag discovery pattern, adapted to KB-OS’s actual task/note schema. Not yet visually verified in-app.
  • Kanban Action Planner (free, MIT license) — its WBS view renders parent/child hierarchy as a collapsible tree with rollups. Potentially closes the exact gap ecco-replacement-research (2026-08-20) said no candidate had. Talbot approved install + test.
  • Dataview Serializer — candidate fix for ecco-view.js’s render-only output being invisible to grep/AI. Talbot approved install + test (revertible).
  • Obsidian CLI (official, free) — flagged as the actual render-verification mechanism (its eval/screenshot commands), since no generic desktop-control tool exists in a Claude Code session.

All follow-on execution (GUI installs + in-app verification, none of which a CLI-only session can do) spun out to Core/Processes/Projects/KB-OS/Tasks/Obsidian-Upgrades-Phase2-Verification.md.

Where findings live

Section titled “Where findings live”
  • Scored candidate table + status: Core/Processes/Projects/KB-OS/UPGRADES.md → “Obsidian Automation Guide — Upgrade Candidates”
  • Focus/Ecco-specific candidates: Core/Processes/Projects/Focus/UPGRADES.md
  • ROADMAP milestone: Core/Processes/Projects/KB-OS/ROADMAP.md → M4.5

Lessons promoted

Section titled “Lessons promoted”
  • No desktop-control tool in Claude Code sessions; Obsidian CLI’s eval/screenshot is the real render-verification path → KB-OS/LESSONS.md
  • Evaluating a paid bundle vs build-it-yourself: check each component at its own source (GitHub/repo), not the sales page → ai-config/AGENTS.md (Hard Rules, after “CHECK RELEASE ASSETS, NOT WRITE-UPS”)