How does the Fabric work within PAI ?

[smolcompute]

I'm confused about the status & workings of the Fabric repo inside of PAI.

Path Questions

From what I understand, Fabric has it hardcoded to look for patterns at ~/.config/fabric/patterns .
That's where the setup process places them. Yet here they come in a skill folder here and the SKILL itself instructs:

# Check if Fabric repo exists
if [ ! -d "$HOME/.claude/skills/fabric/fabric-repo" ]; then
  echo "Fabric repository not found. Cloning..."
  cd "$HOME/.claude/skills/fabric"
  git clone https://github.com/danielmiessler/fabric.git fabric-repo
  echo "Fabric repository cloned successfully."
else
  echo "Fabric repository found at $HOME/.claude/skills/fabric/fabric-repo"
fi

How do I get PAI to use the patterns inside the PAI repo ?

Update questions

The documentation says:

To update patterns:

cd ${PAI_DIR}/skills/fabric/fabric-repo
git pull origin main

This seems to imply that fabric-repo is its own separate repo with its remote set to the Fabric repo. But that's not how things are actually set up. So what's the deal ?

I think the logical thing here would be to ship the fabric-repo as its own repo with upstream fetching from https://github.com/danielmiessler/fabric.git and having PAI's gitignore set to ignore skills/fabric/fabric-repo/. Either that or rely on Daniel to keep fabric-repo/ up to date with the actual Fabric repo.

How have people set this up?

[zestysoft]

I asked an llm to possibly explain what's going on here because I had similar questions -- this answer seems plausible, though not necessary a satisfactory answer to your question about maintenance:

You've spotted a subtle but critical shift in Daniel Miessler's philosophy between Fabric (Jan 2024) and PAI (late 2024/2025).
The reason isn't just messy code management—it’s a deliberate evolution in his thinking about Agents vs. Prompts.
Here is why PAI "absorbs" Fabric's philosophy but re-implements its functionality locally rather than just linking to the external repo:
1. The Shift from "Thinking" (Fabric) to "Doing" (PAI)
*   Fabric is a library of Prompts. It takes text input, applies a "Pattern" (wisdom/structure), and gives text output. It is fundamentally stateless and passive.
*   PAI is a system of Agents. An agent doesn't just "analyze" text; it creates files, runs git commands, speaks to you, and remembers context.
Why the merge?
If PAI just "linked" to Fabric, it would only be able to generate text. By re-implementing them as PAI Commands (~/.claude/commands/), Miessler upgrades them from "Prompts" to "Agent Workflows" that have permission to execute tools (Bash, Write, etc.).
2. The "CLI-First" & "Deterministic" Doctrine
The PAI Constitution explicitly states: "Code Before Prompts" and "CLI-First Architecture."
*   Fabric: Relies 100% on the LLM to do the work via a prompt.
*   PAI: Wants to rely on Code (TypeScript/Python scripts) to do the work, with the LLM just orchestrating it.
The Conflict:
He couldn't just pull in Fabric because Fabric is "Old Way" (Prompt-driven). PAI is "New Way" (Code-driven). He is likely porting Fabric patterns into PAI by wrapping them in executable CLI tools or agent instructions that bind them to specific actions, making them "Deterministic" rather than probabilistic.
3. Context & Identity Injection
Fabric patterns are generic (they work for anyone). PAI is Personal.
*   Fabric Pattern: "Analyze this security log."
*   PAI Command: "Analyze this security log knowing that I am this person, this is a specific repo, and we use these specific security protocols defined in SKILL.md."
By embedding the commands locally in PAI, he can inject the System Context (Voice settings, Directory structure, History logging) that an external Fabric library wouldn't know about.
4. Latency & "The Loop"
PAI is designed to live in the terminal as a high-speed loop (User -> Command -> Tool -> Action -> Voice -> History).
If PAI had to call out to an external Fabric CLI or API for every thought, it introduces friction. By "baking in" the best Fabric patterns as native PAI Commands, the agent has immediate, zero-latency access to those cognitive models without context switching.
Summary
He didn't "partially merge" them to be messy; he is sublimating Fabric.
Fabric was the "Pattern Language" for how AI should think.
PAI is the "Operating System" where those thoughts execute actions.

[smolcompute]

I hope @danielmiessler makes new videos about this since things have changed a lot since his last Fabric video. For one thing, Fabric seems to rely on API keys for intelligence. But I'm guessing most people are using PAI with a Claude subscription.

It's not clear either if you are supposed to install Fabric to use it with PAI or if it's supposed to just work as a skill.

More generally, I hope Daniel makes some tutorials once we get to a version 1.0.

[sti0]

You can use fabric with Claude Code. No API key needed. Just select Antropic within setup process, leave API key empty and select true on OAUTH.

I used fabric with PAI but had do install CLI anyway. So its not clear to me why the fabric patterns exists in this repo, too.

[danielmiessler]

I actually just added new functionality so that you can just use Fabric native as well. So there's actually a Fabric tools directory under core now. And PAI should know how to actually just use it for prompts like super smooth. Yeah, it's pretty easy to just change some prompting and have this work natively, but now it's built in.

…

On Thu, Dec 04, 2025 at 3:17 PM, Timo S. < ***@***.*** > wrote: You can use fabric with Claude Code. No API key needed. Just select Antropic within setup process, leave API key empty and select true on OAUTH. I used fabric with PAI but had do install CLI anyway. So its not clear to me why the fabric patterns exists in this repo, too. — Reply to this email directly, view it on GitHub ( #102 (reply in thread) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAAMLXR2C7UEGFJOGEIJ5VD4AC6IZAVCNFSM6AAAAACMLX6ZXGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKMJWGYZDGOI ). You are receiving this because you were mentioned. Message ID: <danielmiessler/Personal_AI_Infrastructure/repo-discussions/102/comments/15166239 @ github. com>

[smolcompute]

Nice. Looking forward to experiment with it. thanks

[danielmiessler]

Please refer to this major change coming to the repo that should address your issues. If it doesn't please resubmit. Thank you for the work!

#222