Big Idea

The argument in one line.

Claude Code is worth studying not because it is the best harness overall, but because it has pushed specific primitives — script injection in skills, CLAUDE.md file imports, and code-written multi-agent workflows — far beyond what any competitor currently offers.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You already use at least one AI coding harness daily and want to know which Claude Code-specific features are worth adopting or demanding in your current tool.
You are building or maintaining a competing harness (Codex CLI, Cursor, OpenCode, Pyi, etc.) and want a practitioner's feature wishlist.
You use Claude Code but have not explored skills with script injection, CLAUDE.md imports, Workflows, or the full-screen terminal mode.
You are hitting Anthropic usage limits mid-run and need a practical workaround.

SKIP IF…

You are looking for a comprehensive Claude Code review — this video intentionally skips all criticism.
You have no interest in CLI-based agentic coding tools; the screen demos will not be relevant.
You want beginner setup guidance; this assumes fluency with multiple AI coding harnesses.

TL;DR

The full version, fast.

Claude Code has real weaknesses, but it also has a handful of primitives no other harness has matched. Skills with script injection front-load environmental state before the model even starts reasoning, cutting steps and improving reliability. CLAUDE.md file imports let you bridge the CLAUDE.md/AGENTS.md split with a single @-line. Workflows are the biggest differentiator: the agent writes throwaway JS code to define multi-agent phases, prompts, and parallelism — dynamic by construction, not constrained by a fixed schema. The full-screen terminal mode, mid-run account switching, /by-the-way side-chat, /rewind, and mobile remote control round out the list. The argument is not 'use Claude Code' but 'every other harness should copy these things immediately.'

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 →

Chapters

Where the time goes.

00:00 – 02:37

01 · Cold open + sponsor

Sets up the reluctant-endorsement framing ('I've been through it with Claude Code, to put it lightly'), puts on the hat as a visual bit, then runs a Depot sponsor read (Docker CI acceleration).

02:37 – 07:18

02 · Skills: script injection

Walks through a real Repo Explorer skill that uses Claude Code's script injection to list a local cache directory before the model starts reasoning. Argues this is a primitive that should be part of the skills standard and that other harnesses are too cautious to adopt.

07:18 – 11:56

03 · CLAUDE.md imports + claude.local.md + deep links

Shows @agents.md import to bridge the CLAUDE.md/AGENTS.md split. Demos claude.local.md for personal overrides. Discovers and explains claude-cli:// deep links for launching sessions from HTML pages.

11:56 – 13:51

04 · /by-the-way, sub-agents, worktrees

Demonstrates /by-the-way side-chat running a question in parallel while the main run continues. Shows sub-agent navigation panel. Notes --work-tree flag for ephemeral branch work.

13:51 – 24:09

05 · Workflows: code-mode multi-agent pipelines

The centrepiece chapter. Triggers a 15-agent parallel PR audit workflow, reads through the 240-line throwaway JS it wrote (phases, prompts, PRIORS from memory, verdict schemas, dynamic ternaries). Shows the live token burn (~$100/10 min on Fable). Explains why code-written workflows beat schema-constrained ones structurally.

24:09 – 25:40

06 · Mid-run account switching

Demonstrates switching Anthropic accounts mid-workflow. Future tool calls bill to the new account; the running agent is unaffected. Notes this is 'awesome and really abusable'. Aside: Anthropic suspends Fable/Mythos for foreign actors while recording.

25:40 – 27:47

07 · /rewind, /branch, remote control

Quick demos of /rewind (go back steps, but fast-forwarding is hard), /branch (fork the conversation history), and /remote-control (expose the session to claude.ai and the Claude mobile app). Outro.

Atomic Insights

Lines worth screenshotting.

Script injection in skills delivers environmental state to the model at load time, eliminating one full reasoning step before the agent even starts.
CLAUDE.md can import any file with @path syntax — a single @agents.md line makes CLAUDE.md and AGENTS.md identical without symlinks.
claude.local.md is gitignored by design, solving the team-conflict problem when one developer's preferences clash with a shared CLAUDE.md.
Claude Workflows are not a fixed schema: the agent writes its own throwaway JS program — phases, prompts, parallelism — then executes it.
A 240-line workflow program written and run once is the clearest example of code as a step between model runs, not just the output of them.
Code-mode workflows can programmatically filter MCP server output before it hits the context window, cutting token waste that raw MCP flooding causes.
Workflows on Fable burn roughly $100 every ten minutes with eight parallel agents — orchestrating with Sonnet instead drops the cost to something sane.
Mid-session account switching is per API request: a running agent keeps going, but every future tool call bills to the newly authenticated account.
Fable criticises GPT-generated code significantly more harshly than Opus 4.8 did — the same PR that Opus praised, Fable flagged and recommended closing.
The /by-the-way command opens a parallel side conversation without interrupting the main agent run — a UX primitive no other harness had first.
Full-screen alt-screen mode (CLAUDE_NO_FLICKER=1) takes over the terminal completely and restores the previous view on exit — no scroll-buffer pollution.
Running /remote-control hands the active session to claude.ai and the Claude mobile app — imperfect but immediately useful for away-from-desk monitoring.
Anthropic invented the skills standard; their willingness to extend it with script injection is legitimate — the standard should evolve, not calcify.
The real value of watching a live workflow run is seeing the model write prompts that anticipate its own failure modes and instruct sub-agents to avoid them.

Takeaway

What code-mode agents unlock that prompt-mode never can.

WHAT TO LEARN

When an agent writes its own orchestration code rather than following a fixed schema, it gains the ability to structure parallelism, conditional logic, and prompt generation in ways that no pre-built multi-agent framework can anticipate.

Script injection in skills front-loads real-world state — directory contents, process status, environment variables — before the model reasons, eliminating one round-trip and making skills more reliable by default.
Importing other files into CLAUDE.md with @path syntax lets you maintain a single source of truth for agent context while still adding tool-specific overrides in the same file.
A gitignored claude.local.md file solves a real team problem: personal agent preferences (verbosity, tool permissions, formatting rules) can stay out of shared version control without needing a separate workflow.
Code-mode workflow programs are throwaway by design — the agent writes 200+ lines of JS that runs once and is never committed. The value is in the execution, not the artifact.
Pre-filtering MCP server output in code before it reaches the context window is architecturally superior to flooding the model with thousands of raw rows; the agent can write that filter itself rather than relying on the harness to do it.
Mid-session model account switching works because each tool call is an independent API request — the agent's state lives in the conversation, not the auth token, so swapping auth mid-run is safe.
The /by-the-way parallel side-chat is the simplest example of a broader principle: agent runs do not need to be single-threaded from the user's perspective, and UX that surfaces parallelism explicitly changes how people work.

Glossary

Terms worth knowing.

Skills (harness context): Markdown files that scope an AI coding agent's instructions or capabilities for a specific task. Most harnesses support them; Claude Code's implementation adds script injection at load time, which no other harness currently does.
Script injection (skills): A Claude Code feature that executes a shell command embedded in a skill's markdown when the skill loads, injecting the output directly into the model's context before the conversation starts — useful for delivering real-time state like directory listings.
CLAUDE.md / AGENTS.md: Competing context-file standards for AI coding harnesses. AGENTS.md is supported by most tools; CLAUDE.md is Anthropic-specific and adds features like @-file imports and claude.local.md that AGENTS.md does not standardise.
claude.local.md: A gitignored CLAUDE.md variant for personal overrides. Instructions in this file apply only to the developer's own machine, preventing team conflicts when individuals want different agent behaviour than the shared CLAUDE.md specifies.
Workflows (Claude Code): A multi-agent orchestration primitive where the agent writes a JS program that defines phases (audit, rule, verify), sub-agent prompts, and parallelism, then executes that program to spawn and coordinate up to eight parallel agents.
Code mode (workflows): The design principle behind Claude Workflows: rather than configuring a fixed orchestration schema, the agent writes code to structure its own multi-agent runs. This makes the parallelism and prompt logic fully dynamic.
Fable: Anthropic's internal name for their highest-capability model at the time of this video, accessible via Claude Code subscriptions. Equivalent to what is publicly called Opus 4.5 / Claude 4 max-tier in the broader Anthropic model lineup.
Alt-screen mode: A terminal rendering technique that takes over the full terminal viewport without touching the scroll-back buffer. Claude Code's implementation (CLAUDE_NO_FLICKER=1) means closing the app fully restores the previous terminal state.
Harness: In this context, any CLI or desktop application that wraps an AI model with file access, tool calls, and agentic capabilities — e.g., Claude Code, Codex CLI, Cursor, OpenCode, Pyi. Not the model itself.

Resources

Things they pointed at.

01:35productDepot ↗

06:38linkJoel Hook / PySkill interpolation

11:50toolclaude-cli:// deep link protocol

13:20toolCodex CLI /side command

Quotables

Lines you could clip.

02:37

“I am making this video not because Claude Code is better and I want everyone to appreciate how great it is, but because I want everything else to take the good parts so that I don't miss Claude Code when I use other things.”

States the whole thesis in one sentence — shareable as a standalone pull quote→ newsletter pull-quote↗ Tweet quote

23:14

“Code isn't just the output of a model, code is a step between model runs.”

Tight one-liner that defines the code-mode workflow insight; no setup needed→ TikTok hook↗ Tweet quote

23:05

“It wrote 240 lines of code that is entirely throwaway, that is only going to be executed ever once to trigger this workflow. That's fucking awesome.”

Visceral reaction shot, genuine enthusiasm after a long critical arc→ IG reel cold open↗ Tweet quote

21:05

“Opus loved OpenAI code. Fable hates it.”

Punchy model-personality contrast with no context needed→ TikTok hook↗ Tweet quote

The Script

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.

metaphoranalogy

00:00One of those videos I really don't wanna make, but I've been having a lot of experiences that I wanted to share. As you've probably seen, I've been using Fable a lot. Sadly, Anthropic restricts the hell out of your subscriptions, so you kind of have to use it through their interfaces.

00:12In this case, Claude Code. I've been using Claude Code a lot the last couple days, probably as much as I used it back in December and January when I first got addicted to Opus 4.5. I've been through it with Claude Code, to put it lightly.

00:26And while there are lots of things I do not like about it, especially the desktop app, I have come around to a handful and I've even found a bunch of small things that I think Claude Code does better than every other harness in CLI for doing agentic coding. I can't believe I'm doing this, but I think I have to. We're gonna do a video about the good parts of ClaudeCode.

00:47Think I need a hat for this one.

00:52Let's do this. And as a reminder, just because I have a hat and I have convinced a lot of people to spend a lot of money on Anthropic, they are not a sponsor. I don't think they ever would be considering some of the things I've said about this company in the past.

01:05But I do need to get paid, especially after the seven subscriptions that I've activated for myself and my team over the last few days. I'm $1,400 down, so please pardon me for getting a little bit of money from today's sponsor.

01:14Hey, you. You're probably a real dev, right? Like the type that uses Docker?

01:18Well, if you are, you're wasting a ton of time. Almost all Docker users are. At least all of the ones that aren't already using today's sponsor, Depot.

01:25These guys are the kings of Docker. They made it so much faster that it feels unbelievable. Up to 40 times faster for real world use cases.

01:33Their infrastructure builds containers remotely, works for Intel and ARM CPUs so you Mac people are gonna be more than set, and they also built their own fine grained caching for the layers for your Docker images on their own CDN. What this means is that after one dev runs the build, everyone else just pulls the cache instead, making everything absurdly fast.

01:51And this isn't just good for running builds on your machine. This is incredible for agents that are doing work in the cloud, but more realistically, it's incredible for CI, which is why they provide GitHub CI that's up to 10 times faster than GitHub Actions. If you're willing to move to something a little different from GitHub Actions, their CI engine is so much better.

02:08Not only does it handle actual parallelism where it does one step like installs and then has three things running at the same time, which is unbelievably useful when you're doing real CI work, it also runs way faster on much better infrastructure, and most importantly, is accessible via API and CLI, so your agents are able to run your CI in a loop without having to push up the changes.

02:27No more filing PRs just to trigger CI to see what works and then copy pasting the errors back to your agent. Just let the agent run. If this ad was longer than your Docker builds, get that fixed at soydev.link/depot.

02:37So what the hell am I doing making a video about good parts of Claude code? Well, as I hinted at the start, I've been using it a ton and I've noticed a handful of things I actually really like. From cool UI touches to nice features that only exist in Claude Code, to deep integrations with things like their skill management that are actually quite good.

02:55And I'm mostly making this video not to glaze Claude Code to convince everybody to go use it. I'll be transparent with my goal here. I want every other harness to steal all of these things.

03:04I am making this video not because Claude Code is better and I want everyone to appreciate how great it is, but because I want everything else to take the good parts so that I don't miss Claude Code when I use other things. So knowing that, what are those things? First and foremost, I wanna talk about skills.

03:19You might be confused because skills are a feature in pretty much every one of these harnesses now, and they are. They're just markdown files, right? Mostly, yes.

03:28But this is where I start to really like what Code does differently. One of the things Claude Code does really differently with their skills is the ability to execute scripts in them. This might sound insane, right?

03:41Like, just tell the model to go execute the script. But hear me out. There's a lot of use cases where it's nice for the model to know things when it loads the skill instead of having to load the skill and then do things and then do the next step.

03:54Here's an example of a real skill I made that I use all the time. It's Repo Explorer. This is my markdown file that replaced Ben's whole product that he was building, that I was trying to convince him he doesn't need to build that product.

04:07Now he is convinced he understands, but at the time, this was originally a demo, and I've actually found myself using it all the time. The point of this skill is to allow agents to explore implementations of dependencies.

04:18Like if I'm using effect or react or some weird open source project that the model doesn't know that much about, instead of just hunting through the docs and hoping it finds the right thing, this skill allows the model to clone the open source repo into a specific directory and then explore that specifically. It has a pretty simple description.

04:35Clone and inspect external repositories in a reusable local exploration cache. Use the skill when the user asks to explore, inspect, investigate, compare, or answer questions about a repo that may not already be in the current workspace. The actual skill itself is pretty simple.

04:48Use this skill to explore repos without cluttering the active workspace. I have the specific directory on my computer. In my home path, I have .explorerepos, which is the local cache for repos that I can explore.

04:58You can see the cache contents by running this command. It tells you to list the current repositories before deciding which to use. I even call out that certain host support skill injection here.

05:07Otherwise, it should run this script to find what's there. Then it says check whether the repo's already there. If it's not present, clone it, then explore it.

05:16Relatively simple. You might be seeing the first thing that Claude code does that every other harness doesn't that's really nice. The current cache content section.

05:25If the harness is capable, it will immediately make this directory if it doesn't already exist, and then list its contents. This skips a whole step that the model has to do. And if every other harness supported this, I would be able to write this skill entirely differently.

05:41I would just open with the listing of the content, and then I would say right below, if the repo you want is in this directory, just go there. If not, clone it there. It would make this skill half the length, and it would be meaningfully more reliable and resilient.

05:56But instead, I have to write it in two different ways. One where ClaudeCode gets these cool features, and one where every other harness including Codex, Pi, and more don't. Thankfully, there's a Pi extension that does this and I think OpenCode is starting to introduce this as well.

06:09But this is a feature that QuadCode does really well that I think people are too hesitant to introduce. There are real risks with how much people just arbitrarily install skills that the skill could just arbitrarily execute code. We already deal with that with NPM guys, get over your fucking selves.

06:24This is totally fine. If anything, it's better, because the code is in here to be audited by an LLM. It doesn't actually fucking auto run.

06:31It's, yeah. Okay, It does kind of auto run once it's solved, like, get the idea. This is not as unsafe as half the shit we do with NPM.

06:37It's totally fine. And this is not a new thing for me to call out. Two months ago, I already did call this out, and I said that I want other things to be supportive of it, like Codec CLI, PyCursor, etcetera.

06:48I like this pattern a lot. It is genuinely really good, and it's one of those things that should be part of the skills standard, but apparently just isn't.

06:57Some part of me that's frustrated about Anthropic wants to complain that they are moving too far away from the standard, but I'm not going to. Because the standard should move. The standard should improve.

07:06We should take opportunities to do things better, and this is a thing that Claude Co. Does better than every other harness right now. So we should all go embrace that.

07:13And for those looking for the PySkill, Joel made it. Joel Hook slash PySkill interpolation. Joel's a legend.

07:20Apparently, our friend Mario did not like it though, because it was not secure. So know that as well going in. But that is the nature of this.

07:26It's arbitrary execution, but like it's it's not that big a deal. I think people are overreacting. And to their credit, Anthropic did also invent skills, so I think it's totally fair for them to do stuff like this.

07:36I wish other harnesses would embrace the flexibility that ClaudeCode skills have. Next, I wanna talk about the Claude MD for a bit. I have issues here, in particular that the Claude MD and Agent MD are competing standards where everything uses the Agent MD except for Claude, which insists that it has a special file.

07:53I've come around to that a bit because Claude needs different instructions than other things do. It's annoying when I'm using Claude models in OpenCode or Py, but I have come around to some of the cool things you could do in the Claude MD. I will highly recommend that you don't run their slash init command.

08:09It's pretty bad. They have a new interactive version, but your ClaudeMD should be handwritten.

08:15The thing I wanna call out that I think is genuinely really cool is importing additional files. Markdown doesn't have an import standard, because Markdown's not really a standard. It's a vague vibe at best.

08:27So Claude has a solution. Claude MD files can import additional files using at path slash two slash import syntax. Imported files are expanded and loaded into context at launch alongside the Claude MD that references them.

08:40Both relative and absolute paths are allowed. Relative paths resolve relative to the file containing the import, not the working directory. Imported files can recursively import other files with a maximum depth of four hops.

08:52So if you want to intentionally include things like the readme package JSON, etcetera, you can just tag them in. This also means you can do a very convenient thing for those of us who are annoyed that the agent m d and Cloud m d are different. Watch this.

09:04At agents.md. Now my claud m d is my agents m d.

09:10Not with a symlink, not with some weird bullshit, it just is. But more importantly, if I do notice Claude doing some bullshit that my other models aren't, I can put additional stuff here just for Claude. So now I have the full content of my AgentMD as well as my special Claude stuff just in the Claude file without having to do any weird sim linking or managing two different files.

09:31It just, it works. It's good. And if you have other docs in your repo that are useful to the model, you can tag that in here as well.

09:39This is a really nice pattern, especially because it works with things other than Markdown, and I wish more things would copy it. The problem is that AgentsMD has become the standard.

09:48It's the thing everything uses. So if I do this in AgentsMD and it's supported by Codex, it might not be supported by five other things I use.

09:57So by having the ClaudemDB Anthropic special thing, they're able to do stuff like this. They also have a nice override pattern that I actually quite like with claude.local.md. If you add this to your global gitignore, which I would recommend, I'm probably gonna go do that myself, it allows you to have specific instructions for how you want to work that will not affect the rest of your team.

10:17I do love that they have this as a separate thing that allows me to add my preferences without affecting you and how you work. It gets kind of rough in this way where when one person wants one CloudMD and somebody else wants another, git doesn't resolve that well.

10:31This does. This is a good thing that Anthropic has that as far as I know, none of the other harnesses do. I like this.

10:37While I was reading these docs to get this information, I found another really cool feature I didn't know about, which is launching sessions from links. They actually have their own HTTP code here, claud dash c l I colon slash slash that allows you to open things in claud c l I. So you can have a page that someone makes that has a bunch of links in it to trigger claud c l I.

10:57Apparently, Codex also has deep links, which is very, very nice, cool to see. Does this work with app? The Codex app registers the Codex deep link.

11:04Oh, cool. That's actually really nice. It opens an app.

11:07Good shit. I like that. I might even add this to my skill for making HTML plans, where it has suggestions on how to kick off the job, and I can click the Claude CLI link in that HTML page, and it will just open up the terminal automatically.

11:22That's actually really cool. It does lead into the idea that your computer that you do work on is the same computer that your quad code runs on, which honestly hasn't been the best experience for me even just for performance reasons. I've really enjoyed running quad code on other computers on my network, which this will not work for.

11:38But if you are just using one computer with all of your stuff, this is actually a really cool pattern. If you combine this with HTML planning, you can make some really cool UX with this. It even lets you pass the working directory and a GitHub repo, which will be resolved automatically by Claude code if it's seen it before.

11:53If it doesn't, then it will make the session in your home directory instead. Interesting. So far, everything we've talked about isn't actually inside of Claude Code.

12:00It's just features it supports. But I wanna hop into Claude Code itself and show off some things that I actually find quite nice. In order to show the first one, I need to kick off some work.

12:08So I'm gonna do that. Do a quick audit of this work to make sure it's in a good state, then commit, push, and make a PR.

12:18Cool. By now, bet that was taking work it just did to go make a PR for it. But one of the really nice things ClaudeCode has that I haven't seen anyone else do is a special command, slash, by the way, which lets you ask a quick side question without interrupting the main conversation.

12:33Can you tell me a bit about the performance for backups? And now I have a separate chat inside of this chat going while it's still doing the work that I told it to do.

12:46Apparently, Codex has slash side, which is really cool as well. I'm pumped that other things are doing this now. But when I found this feature, it was the only one that had it.

12:53Super nice to just run a thing like this separately. And here we have the info on the performance. And while that was running, we got a bunch of sub agents queued up to explore.

13:04More cool things that Anthropic was first to, but a lot of other stuff now has, is the ability to like navigate through different sub agents it spun up. This isn't the coolest implementation of it though. The coolest implementation by far is workflows, and nothing else has gone quite as far as Anthropic has here.

13:21One other little thing, and this is just one I've been using a lot because I've been stuck in the CLI more, Work trees are not my favorite way to deal with stuff, but for the ephemeral work where I spin up, do it, file PR, and then drop it forever, having a dash dash work tree that actually behaves properly is really nice.

13:37Now I have the repo I was in, but in a sub work tree in the dot clod slash work trees, which has since been get ignored. This pattern's actually been very nice for me, and I've been enjoying it. To just quickly spin up a work tree on some random idea I have.

13:49But what I wanna show off is workflows. There's a couple ways to trigger a workflow. The easiest is to just tell the model to do it.

13:55The most expensive and most reliable option though is UltraCode. As grotesque as that purple gradient it does is, I do often use it to maximize my utilization of my sub.

14:06That said, I'm trying now to use high and x high directly more and just tell it to do a workflow. I'm gonna voice the text because I'm lazy and wanna just explain what I'm doing here at the same time, so let's do that. I wanna audit the open PRs on this project and figure out what their status is.

14:23Which ones have been trumped, which ones don't need to be left open anymore because similar work has been merged, which ones are ready to go, and which ones need a little more before they're worth pushing over the finish line. I wanna use a workflow to break up all of this work. Before you run the workflow, please output the code you're going to use to run it so that we can read through it together.

14:41The last line of there is probably my favorite part, because workflows aren't just a bunch of tool calls the agent can do, they've fully embraced code mode, which is when you don't just tell the LLM to spit out text to trigger specific stuff and keep all of this stuff in its own memory, you let the agent write code to do a thing instead.

14:59Instead of having all your MCP servers that flood you with thousands of rows from your database, you can just have it write code to call the MCP server, get those thousand rows, and then filter it programmatically before it ever makes its way into the context window. While this is running, I wanna point out how I'm running Claude code because I'm not using the standard UI.

15:18I was an early mover to their full screen mode, which is using the alt screen rendering technique in the terminal. So instead of doing weird regexes to replace text in the scroll back history in the buffer, which is how a lot of these terminal apps work, it instead takes over the full terminal. So when I scroll up, I can't scroll to before I ran Claude code, and when I close it, it fully closes and restores my old terminal view.

15:40So if I like, I don't know, l s a here, I have all this stuff. It's my custom command so I don't leak. I run Claude here, and now it's full screen.

15:49And when I close it, I go back to the terminal. Let's assume I'm using the full screen mode or the flickerless renderer.

15:55I found it pretty good. I would recommend switching to it if you're okay with not having traditional scroll buffer. It does get a little annoying when you're SSH ing with TMux, but I just told my agent to go figure that out and fix it, and now I have fully working scroll in TMux over SSH, which has been a lifesaver.

16:11Apparently, you can turn this on by running slash tui full screen in Claude code. But you can also do Claude code no flicker equals one as an environment variable, that's how I have it set. We now have the workflow written.

16:22So here is what a workflow actually is. If you're not familiar, the point of a workflow is to take a bunch of work and break it up into groups of sub agents that have different system prompts, different roles, different things they are working on. It's not just like, okay, let's have six things go off and look at this stuff.

16:40It's a staged workflow that has different steps along the way. So I was gonna tell it this looks good, run it. Be cautious when you tell it to run a workflow though because these burn tokens.

16:50By default, workflows use up to eight agents in parallel, and it will constantly be spinning up new ones as previous work is finished. Gonna wait till this actually starts running so I can open up the workflow and show you.

17:02Here we are. Once the workflow is running, you can go look at it here, and it currently has 15 agents queued, one for every single one of the PRs that is currently open because that's what I asked it to do, to audit the PRs.

17:13Each time it audits one of those PRs, it will decide whether or not it needs to throw that PR over to this new rule phase, and then after there's a third verify phase. It doesn't even necessarily go through the phases in order, it just sets up these sub agents to find some information and then decide if it should go through additional phases or not.

17:32It's a fully dynamic step by step workflow based on what you're asking for. It does burn usage like fucking mad though, so know that going in.

17:41So we're at 9% used right now. And in not much time, that will bump.

17:46Yep. It's already up to 10. And if I refresh again in a minute, it'll be at like 12 or 14.

17:51From my testing, when I had a workflow going, just one workflow with those eight like parallel threads, it was costing about a $100 every ten minutes. Yeah. No going in workflows are expensive with Fable specifically.

18:04But if you use them with cheaper models or you even tell Fable to orchestrate using Sonnet and Opus instead, you can make the usage here way less aggressive. But just those two sentences I said there were enough for it to go up to 12%.

18:18So use with caution. But I wanna read through the code it wrote for the workflow because it's one of the things I think is coolest about it. This is gonna sound contradictory, so hear me out.

18:26You can never be more dynamic than code. I know that sounds crazy, because code is code. It does what it says.

18:32But when the agent can write code, it is effectively building its own custom feature every time it does a workflow. If we were to build workflows as like a core feature that the agent just calls, it would have to be structured such that you tell the model, okay, here are the different phases you can do, here's how you structure it, here's where you put prompts, and we'll go spin up the rest.

18:52But when you tell it to write code, it's now able to structure it however the fuck it wants, and it takes advantage of that. I've seen Fable write workflows that just weren't compatible.

19:01Some of them weren't even valid JS, because it's trying so hard to push the limits of what this feature can do, which I do actually think is kinda cool. So it starts with this meta export, which is the name of the workflow, open PR audit, says what it's gonna do, audit all 15 open PRs, and it has three phases it defines.

19:17One is audit, which is detailed as one read only agent per PR, then the rule is to pick a winner inside of each overlapping cluster, and then verify which is one adversarial checker per disposition. So every time it finds a thing it's not sure about, it will verify that here. And here are all of the PRs.

19:33This is code that isn't actually being used directly for the workflow. It's not exported. It just has this here for its own usage.

19:41This is all info it got from probably running the GitHub CLI, getting this data, and then putting it into this JS file. As each PR here, it says what the title is, whether or not it's draft, whether or not it's mergeable, and when it was updated.

19:53Whether or it's mergeable is like if it's conflicting or not, which is cool that it got that data ahead of time so the agents aren't stuck figuring that out for each of themselves. Then we have by number, which is a mapping of all of the data here so that it's an object that is number key data value, just to make it easier to go through.

20:12Then it actually pulled some stuff from memory, because I've done this specific audit before. So here it had priors where PR thirty five won in a tournament against thirty seven and thirty nine, PR thirty seven lost that same tournament, PR thirty nine supersede thirty five and thirty seven, but it was empirically refuted, all caps refuted.

20:31Chimera, ReferenceMD documents a return type that code doesn't produce. Two disjointing storage runtimes, four or five smoke functions dead, host yeah, it really did not like that PR. That PR was written by five five, so I Fable, it's funny.

20:44Opus four eight glazed the shit out of GPT code. Fable hates GPT code. It's actually really interesting how different they are in this way.

20:53I would say that this is probably the single area I've seen Fable differ the most from Opus. Opus loved OpenAI code. Fable hates it.

21:00Then it has the main logs, this is things that have recently merged, so it has as a reference. Then it is best to group all the different clusters of PRs that are related. So all of these have to do with files.

21:10All of these have to do with white listing. All of these are auth brokers. The rest are all unique.

21:14Cool that they did that for me. Dispositions, these are all the different states it can be in.

21:19The audit schema, this is how it wants the audit to return when the model is done. The ruling schema, this is for the ruler verifier layer.

21:27And then we have the verdict schema, the audit prompt, which is the prompt that is dynamically generated because it's a function that takes in a PR. It also uses the today definition so that the model knows what today is without having to potentially destroy its context by putting that in the system prompt.

21:42It's actually, I haven't read one of these in full before. This is super cool. This whole thing here, actually, is just a function that returns a string.

21:49It's a template string generator that you pass a PR to that has number, title, branch, all of this stuff, to give that to the model as a prompt, because again, it's doing all of this dynamically on the fly, which is super cool. Here it even has a dynamic ternary where it puts different context based on whether or not there are priors for this PR.

22:06Has a different prompt for ruling, which is deciding whether or not this thing should be used or not, and then a verify prompt. You're an adversarial verifier in the Lakebed repo. Read only origin already fetched, do not fetch, checkout or modify.

22:18It knows how to prompt Claude. Historically, I didn't feel like models knew how to prompt themselves. Fable has pushed me way over the line here.

22:24Fable is great at prompting anthropic models. It knows all the ways they screw up and tells them not to ahead of time. The proposed disposition for PR is the data that it fetched from the other agents.

22:34Try to refute this using actual repo state. The GitHub PR view diff checks, get logs, close superseded, close stale, yada yada.

22:42It's a decent prompt. And then we have the results where they call pipeline, which is their magic global helper, as clusters, which are the groups that we talked about before, an async function for auditing every PR member, that's the first stage, and stage two is the next section, stage three is the end part, but you define this all in a pipeline that takes clusters, which is the data as the first argument, and then the functions to define the sub jobs and the sub runs for all of these other sub agents.

23:10It's fucking cool. It wrote 240 lines of code that is entirely throwaway, that is only going to be executed ever once to trigger this workflow.

23:20That's fucking awesome. Although, I will admit with Fable, that is also expensive. We are now up to 24% used in the time that I read that.

23:29Another 12% of my $200 a month, five hour window. When you see everybody talking about agent loops and letting the agent prompt itself, this is a phenomenal example of it. Should you go use this and burn all of your usage yourself?

23:42Maybe. But at the very least, you should look at this and learn from it, realizing that this is where you get really crazy powers with models when you let them write their own code to do additional sub prompting. Where code isn't just the output of a model, code is a step between model runs.

24:00Genuinely really cool. And it's got me thinking a lot more deeply about how to use these things as well, if I'm being honest with y'all. On the topic of usage limits though, I'm burning through them pretty fast.

24:10That's why I now have to use multiple accounts. Let's say I want to reduce my usage on this account right now. If I was near the end and I was scared, I'd have a lot of reasons I'd wanna do that.

24:20We have this workflow already running, so it's going to keep going. But a problem I've had is that when your usage limit gets hit, this run can stop and it's really hard to recover it if possible at all. But what you can do to handle this is make sure you switch accounts before you get there.

24:35But Theo, isn't that going to screw up the run? Isn't that going to, like, break your current stuff? Not from my experience.

24:42As I mentioned, I have that workflow running right now, and I'm gonna switch accounts. I have to hide myself doing this, but I'll show you. And click authorize, login successful.

24:51And again, this agent was already running, but now, all of its usage is going to be routed to this other account. So even though I already had a run going, it will start future turns on the other account's auth because every time it does a tool call, a new API request is made, and once that happens, the new auth is used.

25:11So it will keep using this in a given generation, but when it runs another command, it reads additional data, it does anything that requires execution of code, a new API request will be made, and that one will go through your other auth. This is awesome and really abusable.

25:26While I was recording this, a fun announcement went live that apparently, Anthropic is no longer letting anyone use Fable because the US government told them to suspend Fable and Mythos access for foreign actors. That is unprecedented.

25:41I have a couple more features I wanna show off really quick before I forget. One that a lot of other things have, but I've actually found myself liking the implementation in Claude Code quite a bit, is branch, as well as rewind.

25:52Rewind, I actually use a ton, because if I accidentally send a prompt or get something wrong, I will use it to like go back one or two steps. I do have an issue with it, is that if you do this erroneously, fast forwarding is nearly impossible.

26:05So be careful with rewind, but when you accidentally send something and wanna go back, it is very nice. Slash branch is the better option for when you want to go do something else or take the history you have now and go somewhere else with it. It's very, very nice.

26:21Thankful more things are starting to add that. I definitely recommend it. One more that a lot of people told me to talk about, which I haven't used yet, but I've heard really good things, so I'm gonna try it here, is remote control.

26:31It's a built in feature where I can type slash remote control. So now I've given the ability for this Claude code instance on my real computer to be controlled through claude.ai, the website, or from my phone using the Claude app. Yeah.

26:43I just went to Claude on my phone. I had to go to Claude code, of course. But here, you probably can't see great, and I'm not gonna do anything to fix that.

26:50I can click audit, open PRs, and workflow planning. And this is the exact run that is going on on my computer right now.

26:57And it just hit a Python decoding error with some of the JSON that it got back as a response. Fascinating. Cool though.

27:04I have that on my phone now. I actually do really like that. I will probably use that more when I'm like running a bunch of shit and then out on the go.

27:10I wish they had a better like proper remote control, the whole instance feature like Codex does, but as a quick way to just throw the session externally so you can do it from other things, that's kinda cool. I think that's all I have to say here though. This is a weird reflective experience for me.

27:26There's a lot of things in Cloud Code that are good. I went out of my way to not talk about the bad here, but I might have to in the future.

27:33So if you want me to do a video where I tear apart all of the things that Claude Code does wrong, let me know in the comments. Hope you enjoyed this one. I need to go crash out about Fable being taken away from me.

27:42I am not excited to do that, but until next time. Peace, nerds.

The Hook

The bait, then the rug-pull.

The title is the thesis: a developer who has spent the past month criticising Claude Code sits down, puts on a hat, and spends 28 minutes cataloguing the things it does better than every other harness on the market — not to praise it, but because he wants its competitors to catch up.

Frameworks

Named ideas worth stealing.

03:20concept

Skills: script injection pattern

Embed a shell command in a skill's markdown using Claude Code's @-syntax. When the skill loads, the command runs and its output is injected into context — delivering real-time state (directory listings, env vars, git status) before the model reasons.

Steal forAny skill that needs to know the current state of the filesystem, running processes, or external APIs before acting

08:05concept

CLAUDE.md @-import bridge

Add @agents.md as the first line of CLAUDE.md to import its full content, then add Claude-specific instructions below. Single source of truth with a safe override layer.

Steal forAny team using both Claude Code and other harnesses who want one context file without symlinks

14:45model

Code-mode workflow structure

Meta export (name, description)
Phase definitions (audit / rule / verify)
PR/task data as constants
PRIORS from memory
Dynamic prompt generators (template string functions)
Pipeline call with cluster groups and async stage functions

The pattern Claude writes when asked to run a workflow: a throwaway JS file that defines the orchestration as code, not as a config schema. Each phase gets its own prompt function; the pipeline helper handles sub-agent spawning.

Steal forDesigning multi-agent orchestration systems; arguing for code-over-schema approaches to agentic pipelines

CTA Breakdown