Capture Concept is an Obsidian knowledge-distillery skill for turning daily learning into interlinked, reusable vault notes.
It helps an AI agent capture concepts, decisions, debugging lessons, meeting knowledge, feedback, aha moments, reading notes, and domain learning maps with the right structure, depth, tags, backlinks, MOCs, and provenance.
It is built for Obsidian users who want AI-assisted personal knowledge management, agent-readable Markdown notes, reusable learning maps, and a vault that compounds into useful context for future AI work.
It ships as an agent skill with ready-to-use adapters for Claude Code and Codex, so you can install it into your AI coding agent and start capturing in minutes.
I started using Obsidian more seriously while mentoring people who were learning many different things at once: software engineering, AI, product thinking, career growth, debugging, system design, and the everyday judgment that usually stays trapped inside conversations.
At first, I was simply sharing useful notes with mentees. Then a pattern became obvious: the best learning moments did not all look the same.
Sometimes someone fixed a bug and needed a reusable postmortem. Sometimes they made a decision and needed to preserve the tradeoffs. Sometimes a reviewer gave feedback and the real value was the skill gap behind the comment. Sometimes an article introduced five concepts, but only two were worth turning into durable knowledge. Sometimes a sentence in a casual conversation became an aha moment.
So Capture Concept was built around a simple idea:
Knowledge should be captured in the shape it naturally has.
The skill grew out of daily mentoring habits. We used it internally to turn interesting concepts into structured Obsidian notes, then reused those notes for study, career growth, and AI context. Over time, the vault became more than a pile of summaries. It became a working knowledge base: searchable by humans, linkable in Obsidian, and reusable by AI agents building new applications.
This repository packages that workflow so other Obsidian users can build the same kind of compounding knowledge base.
- Detects the capture scenario before writing.
- Extracts only concepts that are worth keeping.
- Chooses the right note type and template.
- Calibrates depth as
reference,working, orcomprehensive. - Adds expert enrichment: mechanisms, examples, gotchas, tradeoffs, and sources.
- Generates Obsidian YAML frontmatter, tags, wikilinks, MOCs, and logs.
- Scores backlink candidates and maintains graph hygiene.
- Creates stub candidates for important concepts that do not exist yet.
- Initializes and checks your Obsidian setup with a built-in doctor command.
Capture Concept is useful if you:
- Use Obsidian as a personal knowledge base, second brain, or team learning vault.
- Want AI agents to write notes that are useful after the chat window is gone.
- Capture debugging lessons, architecture decisions, review feedback, meeting insights, or technical reading.
- Reuse Markdown notes as context for LLM apps, RAG systems, prompts, or mentoring material.
- Prefer structured knowledge over generic summaries.
Capture Concept treats knowledge capture as multiple daily-life workflows, not one generic summarization task. The point is not to save everything. The point is to notice what kind of learning moment just happened, then preserve it in the shape that will be useful later.
| Code | Flow | Everyday moment it protects |
|---|---|---|
| S1 | Post-conversation distillation | You just had a long chat with an AI, mentor, teammate, or friend, and somewhere inside it are three ideas you do not want to lose. |
| S2 | Goal-driven concept mapping | You are about to build something, change roles, start a project, or chase an ambition, and you need to know what concepts must be learned first. |
| S3 | Comparative source analysis | You are staring at two articles, designs, code snippets, tools, or answers and thinking, "Why is this one better?" |
| S4 | Build-before-you-build recon | Before spending a weekend reinventing something, you want to know what already exists, what is mature, and what is worth adopting. |
| S5 | Debugging postmortem | You finally fixed the bug, but the real treasure is the failure pattern, diagnostic path, and prevention lesson future-you will need. |
| S6 | Decision journal | You chose a tool, architecture, job move, product direction, or tradeoff, and you want the reasoning to survive longer than your memory. |
| S7 | Reading synthesis | You read an article, paper, book chapter, or URL and want the reusable ideas, not just a pretty clipping in your vault. |
| S8 | Meeting crystallization | A meeting sounded ordinary, but someone revealed a heuristic, mental model, or team principle that should become shared knowledge. |
| S9 | Feedback -> skill gap | A reviewer, manager, mentor, or customer gave feedback, and the useful part is the skill gap hiding underneath the comment. |
| S10 | Aha-moment capture | Something clicked while walking, coding, reading, or talking, and you need to preserve the spark before it gets over-explained. |
| S11 | Workflow reverse-engineering | You watched someone do something well and want to capture the repeatable process, decision points, and variations. |
| S12 | Domain onboarding | You are entering a new field and need a learning map: vocabulary, dependencies, debates, and what "good enough" means first. |
These flows came from the way real learning actually arrives: in debugging sessions, review comments, meetings, rabbit-hole reading, decisions, sudden insights, and the quiet pressure of needing to grow into the next thing.
See docs/CONCEPT_FLOWS.md for the routing logic and depth review.
Required:
- An Obsidian vault directory containing
.obsidian/ - Python 3
- Filesystem read/write access to your vault
- An AI-agent environment that supports local skills (Claude Code or Codex)
- Required companion agent skills:
obsidian-cli,obsidian-markdown,obsidian-bases
Recommended:
rgfor fast vault searchdefuddlefor URL capture- An Obsidian command/URI setup for opening and searching notes
The required companion skills are included in this repository. Platform-specific adapters install them into each agent environment. Capture Concept does not silently install external commands. The init doctor detects what exists and marks setup incomplete until the required companion skills are present.
After cloning the repository:
cd capture-conceptChoose a platform adapter:
- Claude Code: for_claudecode/README.md
- Codex: for_codex/README.md
For Claude Code, one command installs the skill and its companions and walks you through choosing your Obsidian vault:
python3 for_claudecode/scripts/install_claude.py --initPreview the install plan first with --dry-run.
Try a capture:
Capture this debugging fix as a reusable postmortem in my Obsidian vault.
The skill auto-activates from its description. To invoke it explicitly, use
/capture-concept in Claude Code or $capture-concept in Codex.
For full setup details, see docs/SETUP.md.
Capture what I learned from this conversation into my Obsidian vault.
Capture this bug fix as a reusable debugging postmortem.
Save this decision and the tradeoffs behind it to my Obsidian vault.
Capture this aha moment before I lose the original framing.
Create a domain learning map for someone new to retrieval augmented generation.
Golden examples live in examples/.
.
├── README.md
├── docs/
├── examples/
├── for_claudecode/ # Claude Code adapter
│ ├── README.md
│ └── scripts/
│ ├── install_claude.py
│ └── install_claude.sh
├── for_codex/ # Codex adapter
│ ├── README.md
│ ├── scripts/
│ │ ├── install_codex.py
│ │ └── install_codex.sh
│ └── skills/
│ └── ... Codex agents/openai.yaml overlays
├── scripts/
│ ├── validate.sh
│ └── validate_skill.py
└── skills/ # shared, platform-agnostic skill tree
├── capture-concept/
├── obsidian-cli/
├── obsidian-markdown/
└── obsidian-bases/
Claude Code reads each skill's name and description directly from its
SKILL.md frontmatter, so the Claude adapter needs no metadata overlay. Codex
uses agents/openai.yaml overlays, which live under for_codex/.
See docs/COMPANION_TOOLS.md for required companion skills and recommended command-line helpers.
Short version:
rgmakes vault search faster.defuddleimproves URL-to-Markdown extraction.- Obsidian URI/command support can improve note opening and search workflows.
obsidian-cli,obsidian-markdown, andobsidian-basesare required companion skills and are included in this repo.
Capture Concept writes to the configured Obsidian vault. It does not upload your vault by itself. URL capture may fetch external web pages when you ask it to capture from a URL.
Security posture:
- The installers preview their plan and ask before writing files.
- The Claude Code and Codex installers only install the four reviewed skill directories in this repo (plus the Codex metadata overlay for Codex).
- Installs are non-interactive only with
--yes; existing installs are never overwritten without--replace. - Symlinked package content, generated caches, and unexpected skill folders are rejected by validation, and installs do not dereference symlinks.
- Existing installs are backed up outside the active skills directory.
- Config is written with private file permissions and symlink config writes are refused.
- Helper writes are constrained to the configured Obsidian vault.
- Doctor output hides local command/skill paths by default; use
--show-pathswhen debugging setup.
Read docs/PRIVACY.md before using this on private or synced vaults.
Validate locally:
./scripts/validate.shPlatform-specific development and update commands live in their adapter folders. See for_claudecode/README.md for Claude Code and for_codex/README.md for Codex.
Contributions are welcome. See CONTRIBUTING.md for local validation, contribution scope, and review expectations.
Version: see VERSION.
This project is early but usable. It is built from a real mentoring workflow and is being open-sourced so other Obsidian users can adapt it for learning, work, career growth, and AI-context knowledge bases.
MIT. See LICENSE.
Obsidian, Obsidian.md, Obsidian plugin alternative, Obsidian AI, AI notes, AI agent skills, Claude Code skill, Claude Code Obsidian, Claude skills, Anthropic Claude, Codex skills, OpenAI Codex, knowledge management, personal knowledge management, PKM, second brain, Zettelkasten, Markdown notes, note-taking, learning tools, mentoring, RAG context, LLM context, technical writing, debugging postmortems, decision journals, meeting notes, reading synthesis.