obsidian-second-brain

obsidian-second-brain is a cross-platform skill that turns an Obsidian vault into persistent memory for CLI coding agents. The README states it runs on Claude Code, Codex, Gemini CLI, OpenCode, Antigravity, Hermes, Pi and Grok Bot, with the vault itself acting as the shared memory layer so context survives switching between tools. The stated problem is that agent sessions start from scratch and notes in Obsidian sit unused, so the project connects the two: agents read and write plain markdown notes in the vault. Command surface (README lists 47 commands, 46 in builds without the Google Calendar MCP): - Operations: /obsidian-save, /obsidian-ingest (URL, PDF, audio, screenshot), /obsidian-synthesize, /obsidian-reconcile, /obsidian-export, /obsidian-daily, /obsidian-calendar (agenda, reconcile, meeting, schedule), /obsidian-recurring, /obsidian-log, /obsidian-task, /obsidian-person, /obsidian-capture, /obsidian-catchup, /obsidian-find, /obsidian-recap, /obsidian-review, /obsidian-board, /obsidian-board-hygiene, /obsidian-project, /obsidian-projects, /obsidian-health, /obsidian-merge, /obsidian-reindex, /obsidian-retrieval-eval, /obsidian-decide, /obsidian-visualize, /obsidian-learn, /obsidian-init, /obsidian-architect, /create-command. - Thinking tools: /obsidian-brainstorm, /obsidian-challenge, /obsidian-panel, /obsidian-emerge, /obsidian-connect, /vault-deep-synthesis, /obsidian-distill, /idea-discovery, /obsidian-graduate. - Context: /obsidian-world, described as loading identity and state with progressive token budgets (L0-L3). - Research: /x-read, /x-pulse, /research, /research-deep, /notebooklm, /youtube, /podcast. These use xAI Grok for live X access, Perplexity Sonar for web research, and free key-less sources (Wikipedia, HackerNews, arXiv, Reddit and others) when no key is configured. YouTube transcripts come via youtube-transcript-api, with optional metadata and comments through the YouTube Data API v3; an optional --visual mode extracts scene-change frames with ffmpeg for vision reading. Notable design claims in the README: notes are written "AI-first" with a "For future agent" preamble and frontmatter for retrieval; an OKM (Open Knowledge Metabolism) freshness policy classifies facts as timeless, dated or pointer; bi-temporal facts record when something was true and when the vault learned it; typed edges in a relations graph are linted; and a tag taxonomy audit is opt-in via _meta/taxonomy.md. The project describes itself as an evolution of Karpathy's LLM Wiki pattern, differing by rewriting existing pages rather than appending, auto-resolving contradictions, and running scheduled agents. Scheduled/background behavior: four scheduled agents are described (morning brief, nightly consolidation, weekly review, vault-health check), with the nightly agent running five phases including reconciliation, synthesis, orphan healing and index rebuild. Other details: MIT license; v0.15 "The Port" adds Windows support (paths, CRLF/BOM handling, UTF-8, USERPROFILE-aware config home), Gemini-first /podcast with a Whisper fallback, Simplified Chinese triggers, /obsidian-reindex, an Obsidian lint plugin, a reproducible retrieval benchmark, a generated docs site, typed edges and a callout preamble; the README mentions a 763-test CI wall. A Telegram bot capture flow is referenced via /obsidian-catchup. The README also links demos, a changelog, discussions and the author's blog. This overview reflects only what the repository README states; no independent verification of behavior, quality or performance was performed.