How I Use Obsidian + Claude Code to Run My Life

Summary of the Video (Obsidian + Claude Code Workflow)

The video explains how the speaker uses Obsidian (a “second brain” built on Markdown files) together with Claude Code (an agent controllable via the command line) to run personal planning, reflection, and idea-generation workflows. The core claim is that the system becomes “game-changing” when Claude Code is given the right, structured context from an Obsidian vault—especially using Obsidian’s link graph and CLI integration.

Core concepts

• Claude Code: A command-line agent that can perform computer actions using natural language (e.g., create/open files).
• Problem addressed: Without good context, you waste time re-explaining projects to agents; web-based chat “memory” isn’t transparent or controllable.
• Obsidian vault: A directory of Markdown files (“vault”) where notes are connected via backlinks/links, forming an interrelationship graph that mirrors how people think.

---

Key Product/Feature Ideas

1) Context injection via files (instead of repeating explanations)

Claude Code can reference a project description/context file stored in the vault. When starting new sessions, the user can pass that file so Claude Code already “knows the setup,” reducing repetitive prompting.

2) Obsidian CLI as the missing bridge (vault + relationships → agent understanding)

Obsidian CLI (a tool mentioned in the video) allows Claude Code to:

• read vault files
• receive metadata about inter-file relationships (the link graph)

This helps Claude Code infer patterns such as:

• latent themes that repeat across months
• contradictions/counterevidence in the user’s historical notes
• which topics are central vs. orphaned/unlinked

3) Slash commands for personal automation (agent workflows)

The video demonstrates an environment of pre-built “/commands” run through Claude Code, including:

• /context: loads full situational context (daily notes, project notes, backlinks) before work begins.
• /today: creates a prioritized daily plan using calendar + iMessage/tasks + relevant daily-note history.
• /close day: end-of-day processing extracting action items, surfacing vault connections, and adding “confidence markers.”
• ghost: answers like the user; builds a “voice profile” from the vault, writes, then evaluates fidelity.
• Challenge topic: pressure-tests beliefs using the vault to find contradictions/counter evidence and shift thinking.
• Emerge: surfaces implied ideas/patterns the user never explicitly stated.
• /drift: compares “stated intentions” vs. actual behavior over 30–60 days.
• Deep 30-day vault scan: cross-domain pattern detection + graph analysis to generate ideas (including tools/products/films).
• Trace: tracks how an idea evolved over time using graph traversal and timeline reconstruction.

4) Commands can be generated/created by the agent

The video describes a loop where Claude Code can propose commands and even generate “demo” versions to avoid exposing sensitive personal information during public demos. Commands can also be “built” iteratively based on higher-level goals (e.g., “make a command that figures out daily focus” rather than crafting everything manually).

---

Guides / Tutorial-Style Workflows Highlighted

A) “Move from reflection to action” via idea promotion

The system distinguishes between:

• ideas in daily notes
• ideas that become standalone notes that can compound via backlinks

A proposed/previewed workflow includes a command like /graduate:

• scans recent daily notes for ideas (tagged or not)
• helps the user decide whether to “promote” them into a new or existing note
• adds backlinks so the idea grows inside the graph

B) Idea generation across the vault (comprehensive context gathering)

The video shows an “ideas” run that:

• gathers vault structure
• detects orphans/unresolved links and tag counts
• reads daily notes and specific context files (e.g., project/workflow/personal show context)
• outputs an “idea generation report” with:
  • “structural highlights”
  • “actionable” next steps (tools to build, systems to implement, subjects to investigate)

C) Inline delegation / task triggers

Example scenario: start a scheduling interaction (“schedule a meeting?”) where the agent checks the calendar and daily notes to decide whether a meeting is redundant because existing commitments already cover it.

Suggested UX pattern:

• delegate tasks directly from notes (e.g., “start small—write a schedule request in a daily note; let an agent handle the rest”)

D) “Connect” command (graph-based concept linking)

A /connect command connects two domains by exploring their neighborhood in the vault graph.

Example:

• connecting “filmmaking” and “worldbuilding,” then producing conceptual “bridges” by linking themes across notes and essays.

E) Voice/persona + fidelity checking

Ghost demonstrates writing in the user’s “voice profile,” then evaluating fidelity against the vault.

---

Analysis Claims / Outcomes Emphasized

The speaker argues output quality depends heavily on whether Claude Code has:

• the right, up-to-date context
• enough vault data to connect dots

The system is described as enabling “breakthrough” insights such as:

• latent patterns
• contradictions in beliefs
• ideas implied by scattered premises

Tradeoffs:

• processing is slower because Claude Code reads many files
• privacy risk is significant when giving agents access to personal information

---

Privacy and Safety Theme

The speaker repeatedly emphasizes:

• giving agents large amounts of personal data is “scary”
• creating demo versions of commands during public demos
• the approach raises questions about how privacy norms evolve

---

Main Speakers / Sources

• Vin (Internet Vin) — main guest/source demonstrating the workflow and commands.
• Greg Eisenberg — host and main narrator asking guiding questions and responding throughout.