14:38Stephen G. Pope · Tutorial
How to Build a Coding Agent like Claude Code
A 14-minute ground-up build of a CLI coding agent that demystifies what Claude Code actually is under the hood.
May 24thThe argument in one line.
Embedding an AI agent directly inside a knowledge management app eliminates the constant context-switching between notes, terminal, and AI chat that turns most second-brain setups into two jobs running at once.
Read if. Skip if.
READ IF YOU ARE…
- You already use Obsidian and are frustrated that your AI tools live in a separate window.
- You manage API secrets, custom agent skills, or team workspaces and want them unified in one free, self-hosted tool.
- You want to build and deploy custom AI skills from within your note editor without touching a terminal.
- You need cross-device or team sync without paying for an external sync service.
SKIP IF…
- You want a non-technical, drag-and-drop notes app — Shockwave requires comfort with GitHub and API keys.
- You are not interested in running or extending open-source software.
The full version, fast.
The friction in most AI-plus-notes workflows is not the AI itself but the constant switching between apps. Shockwave is a free Obsidian clone that bakes an AI coding agent directly into the editor: select text, prompt the agent, and the result appears inline. Skills are installable from any GitHub repo, secrets are managed inside the app, and new skills can be generated from API docs in a single prompt. A GitHub sync layer handles cross-device and team workspace sharing with no paid external services required.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:20
01 · Intro to Shockwave
Hook and feature promise: free, open-source, AI agent built in, GitHub sync.
00:20 – 00:37
02 · Notes, Links & Graph View
Obsidian-parity demo: sidebar notes, internal links, backlinks, and graph view.
00:37 – 01:08
03 · Integrated AI Agent Setup
Settings walkthrough: provider, model, API key, system prompt, global/workspace skills, secrets manager.
01:08 – 01:53
04 · Using the Agent in the Editor
Select text, prompt agent to list top coding agents in 2026, result inserted inline. Follow-up: 'be more concise' — agent modifies in place.
01:53 – 02:43
05 · Downloading & Using Skills
Clone a marketing skills GitHub repo, load content-strategy skill, agent walks through a content strategy interview.
02:43 – 03:59
06 · AI Thumbnail Generation
Paste YouTube URL, agent fetches thumbnail, Gemini analyzes it, Key API regenerates with the creator swapped in, result drops into the document.
03:59 – 05:32
07 · Building New Skills
Agent reads Scrape Creators API docs and secret, builds a new YouTube report skill, self-corrects two errors, then fetches top 10 Claude Code videos into a document.
05:32 – 07:10
08 · GitHub Sync Across Devices
Create a GitHub repo from Shockwave, auto-sync files, live two-way sync demo — edit on GitHub, see change appear in Shockwave instantly.
Lines worth screenshotting.
- The bottleneck in AI-assisted note-taking is not intelligence — it is tab-switching; collapsing everything into one surface removes the friction that kills momentum.
- A skills system turns a general-purpose AI agent into a specialized one; the same model behaves differently the moment it has domain instructions loaded.
- Agent self-correction is underrated: the skill-builder in this demo failed twice, auto-fixed, and succeeded without any human intervention.
- GitHub as a sync backend costs zero and gives version history, branching, and team access in one move.
- Secrets management inside the tool is the missing feature that stops most people from sharing agent workflows with their team.
- Inline editing — where the agent writes directly into the document at the cursor — is a fundamentally different UX from copy-pasting from a chat window.
- Giving the agent access to external API docs lets it build its own tools on demand; the skill becomes a reusable artifact, not a one-off prompt.
- Open-source second-brain tools with embedded agents may erode Obsidian's moat faster than any paid competitor could.
What embedded agents actually fix in a notes workflow.
WHAT TO LEARN
The problem with combining AI and notes is not the AI — it is the gap between where you think and where the AI lives.
- Context-switching between a notes app, a terminal, and an AI chat window is where most workflows break down; unifying them in one surface eliminates the friction, not the capability.
- Loadable skills let the same AI agent specialize without retraining — a content-strategy skill, a thumbnail-generation skill, and a YouTube-data skill all run through the same model but behave like distinct tools.
- Giving an agent access to API documentation and secrets lets it build its own integrations on demand; the resulting skill is a reusable artifact, not a one-off prompt that disappears after the session.
- Agent self-correction is a meaningful reliability feature: when the YouTube skill failed twice on first run, the agent diagnosed and fixed the error without human input before executing successfully.
- Using GitHub as a sync backend costs nothing and adds version history, branching, and team access as a side effect — no external sync service required.
- Inline editing — where output appears directly at the cursor in the document rather than in a chat panel — changes how you interact with AI output, from paste-and-fix to generate-and-continue.
- Secrets management inside the tool is what makes sharing agent workflows with a team practical; without it, every collaborator has to manually configure their own credentials.
Terms worth knowing.
- Shockwave
- A free, open-source desktop note-taking app modeled on Obsidian, built with a native AI coding agent that can read and write files directly inside the editor.
- Skill (agent skill)
- A structured prompt or script that defines how an AI agent should approach a specific task. In Shockwave, skills are loaded from GitHub repos and made available to the agent as callable tools.
- Inline editing
- The agent writes its output directly into the document at the selected location, rather than returning it in a separate chat panel for the user to copy.
- GitHub Sync
- A built-in feature that commits and pushes workspace files to a GitHub repository automatically, enabling cross-device access and team collaboration without a paid sync service.
- PI coding agent
- The embedded AI agent runtime used by Shockwave. Supports tool use, file modification, secret access, and skill loading.
Things they pointed at.
01:53toolMarketing Skills Repo (GitHub)
02:43toolGemini (image analysis)
02:43toolKey API (image generation)
03:59toolScrape Creators API
Lines you could clip.
01:55
“Makes it a lot nicer than having to move to different apps or going to your terminal and running these different things in Claude Code and trying to get it to integrate with these files somewhere else.”
05:30
“You don't have to pay for any external services.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
00:00Today, I'm gonna show you guys the new Obsidian clone that I built that has a fully integrated AI agent that can modify files, use skills, access APIs, and it's totally free, open source, and it also integrates with GitHub so that you can switch from one computer to another and all your files are there, or you can use it to share different workspaces with your team.
00:17Let's get into it. So when you open up Shockwave, it's gonna look a lot like Obsidian where you have all of your notes on the left hand side. You can create links to different pages.
00:26You can follow those links, and then all of the other pages that link to that page will be referenced below. And of course, you've got the graph view where you can see how all of your pages are interlinked between them. But the thing I wanna show you today is they fully integrated agent chat using the PI coding agent that is built directly into the app.
00:43So when you go into your settings, you have AI features built right in. You've got the agent chat, you can pick whichever provider you want, the model, put in your API key and then it has the system prompt which you can modify that explains to the coding agent how to interact with the system and the note system so that it can also modify those files.
00:59You can also upload global skills and workspace level skills and it also manages all of your secrets. So now to show you the power of that, if I open up a new file, coding agents, from within the editor I can create a header, the best coding agents, and I can select that text, I can message the agent, and then down here I can just say list the top 10 coding agents in 2026.
01:26Go ahead and run that. You'll see it executing on the right side including any tools that it's using and then it will just insert that directly into the document.
01:34You can always modify it. Send to agent. Please be more concise.
01:40And then it just makes those modifications directly in line with your editor, makes it a lot nicer than having to move to different apps or going to your terminal and running these different things in Cloud Code and trying to get it to integrate with these files somewhere else. But beyond that, have access to skills so you can download skills.
01:56This is a marketing skills repo that just has a bunch of different marketing skills to add to any agent. I'm going to just download that repo, come back to Shockwave, go to skills, choose a folder, you can just open it up, jump into the marketing skills, list out all the tools.
02:11Let's say I wanna grab a skill for content strategy, you just add that in there. We'll come back and reset the chat. What skills do you have?
02:19And now we see we've got the content strategy. So you could just come here, content strategy, message agent, help me build a new content strategy.
02:31And then it's just gonna walk you through whatever that skill is programmed to do to help you build out that content strategy. My guess is it's gonna ask you some questions and there it goes. It starts to ask a bunch of different things.
02:41And let's say I wanted to grab a thumbnail I could use for myself. I'll go ahead and use this one because this is what my app literally helps you replace. You can just grab the link.
02:49We'll paste in the YouTube URL, clear the chat, grab the YouTube thumbnail for this video and drop it below.
03:03It'll pull it right up. Then I can drop my own image below and I can just select the two thumbnails, message agent, take the top thumbnail and replace the person with the image below keeping everything else the same except change system to testing.
03:30So I'm using some of the skills that I wrote which are also accessing these API secrets. It's using Gemini to analyze and then it's using Key API to regenerate the thumbnail and there we go.
03:42So you can see it's like a complete copy down to every last little piece except we got me built in here instead. So you can see how I've integrated the AI agent to call skills and actually integrate with the document editor on the fly.
03:54By the way, if you're enjoying this video, make sure to like and subscribe. It tells me what type of content you want more of. And building new skills from within the system is easy as well.
04:01Let's say I wanted to get a YouTube report for all the recent YouTube videos. I could use this API that I've been trying out, scrape creators. I could come back here, just add this secret.
04:15Drop in the token, create me a new skill for getting the top 10 most recent viral videos from YouTube on Claude code and I could just pass it the docs.
04:34I wouldn't even really need to do this but makes it a little faster using the API key from the secrets.
04:43So from here, it'll actually go and get the secrets, pull the one that needs, look at the API docs, build the skill and then make it available to me so that I can use it here and I can pull reports for YouTube. Alright. So it looks like we've got the skill created.
04:59Then I can message the agent. So here it just inserts exactly where I want it to drop the information in what file. Give me the top 10 Claude code videos from YouTube.
05:14And you can always expand these to see what's going on. Looks like the first couple times that it tried it got an error so it was just fixing the skill and then here we go. We've got the top 10 videos.
05:25By the way, if you wanna see me build this stuff, I go live every day at 10:00 Pacific time. Make sure to check out my Twitch channel. I'll see you there.
05:31Now the one other thing that was important to me was making sure that I had a sync with GitHub so that if I ever built one of these workspaces and I wanted to move from one computer to the other or if I wanted to, you know, work with my content team so we could all share the same workspace, I wanted it to work with GitHub and the benefit of that as well is that you don't have to pay for any external services.
05:50So you just put in your API key to GitHub, jump over to your workspace. This is the active one.
05:57We go to sync. We're gonna create a new GitHub repo. We'll call it demo.
06:02We'll create that repo. This will actually create it. So you can see here it already automatically synced all of the files just a second ago up to GitHub.
06:11If I make a change here to the GitHub report new, you'll see it sync down here below. This will actually link back to GitHub as well and you can see it was just changed to YouTube report new.
06:23And if I change this back, edit it directly on YouTube, just pretend we're editing it from some other computer in Shockwave. YouTube report new new Commit changes.
06:37We'll see the sync and we got new new. Bam. And if you want to learn how to build out this type of software where you integrate AI agents directly into your own custom software, I walk through how to build this app step by step inside the AI architects.
06:50I'm also building out a new course that will take you from a complete beginner to a expert AI product engineer. I'm also dropping my daily lives here as well and I break them up into different pieces so it's easy to watch. And it's the only place you can actually watch the replays.
07:04I'd love to see inside the community. Either way, I hope you enjoyed this video and I'll see you on the next one.
The Hook
The bait, then the rug-pull.
What if your notes app could also write the notes? Stephen G. Pope opens on the bold claim then immediately delivers: Shockwave, his free open-source Obsidian replacement, ships with an AI agent that edits documents inline, loads skills from GitHub repos, manages API secrets, and syncs the whole workspace to GitHub.
Named ideas worth stealing.
01:53concept
Skill-as-tool pattern
Agent skills are loaded from GitHub repos as structured instruction sets. Any skill installed becomes a callable tool the agent invokes automatically when the task matches.
Steal forany AI agent workflow where you want reusable, shareable, version-controlled prompts
03:59model
In-context skill generation
Give the agent API docs plus a secret key, ask it to build a skill. The agent writes the skill code, stores it, and makes it immediately callable — no manual coding required.
Steal forrapid integration of new APIs into an agent workflow without writing boilerplate
How they asked for the click.
VERBAL ASK
06:09product
“If you want to learn how to build out this type of software where you integrate AI agents directly into your own custom software, I walk through how to build this app step by step inside the AI Architects.”
Soft sell at the end. Mentions community, a beginner-to-expert AI product engineering course, and daily live replays. Non-aggressive, product-first positioning.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
