I've been using Publii to manage my site but apparently my site is too big for GitHub anymore. Any time I try to update my site and upload to GitHub, I get a 502 error.

Pretty frustrating. That GitHub endpoint times out because there's simply too many files for it to handle. I've enjoyed having my website in a private GitHub repo so I could have a backup of my website before it's eventually published on Cloudflare. However, I already have everything backed up in Google Drive. Additionally, Cloudflare keeps a history of every time I deploy an update to my site.

It's time to cut GitHub out of the equation. Here's how I've reconfigured my Publii/Mac setup so I can publish straight to Cloudflare without changing my workflow.

Reconfigure Publii to publish locally

Changed the Server Type to Manual and use a different output directory.

Create a Cloudflare token

This is for a local script to deploy it once it detects changed in that directory. I need to authenticate via an account token.

Create a bash script to deploy to Cloudflare

#!/bin/bash
DEPLOY_DIR="/Users/mattsayar/publii_deploy/mattsayarcom-files"
PROJECT="publii-cloudflare"
LOCKFILE="/tmp/publii-deploy.lock"

export HOME="/Users/mattsayar"
export USER="mattsayar"
export CLOUDFLARE_API_TOKEN="api_token"
export CLOUDFLARE_ACCOUNT_ID="account_id"

cd "$HOME"

/opt/homebrew/bin/fswatch -o -l 15 -e "\.DS_Store" "$DEPLOY_DIR" | while read; do
    NOW=$(date +%s)
    LAST_DEPLOY=$(cat /tmp/publii-deploy.last 2>/dev/null || echo 0)
    ELAPSED=$((NOW - LAST_DEPLOY))
    if [ "$ELAPSED" -lt 60 ]; then
        echo "[$(date)] Last deploy was ${ELAPSED}s ago, skipping (cooldown)."
        continue
    fi
    if [ -f "$LOCKFILE" ]; then
        echo "[$(date)] Deploy already running, skipping."
        continue
    fi
    touch "$LOCKFILE"
    echo "[$(date)] Change detected, deploying..."
    wrangler pages deploy "$DEPLOY_DIR" --project-name="$PROJECT" --branch=main
    echo "[$(date)] Deploy finished."
    date +%s > /tmp/publii-deploy.last
    rm -f "$LOCKFILE"
done

Don't forget to chmod +x deploy.sh. The fswatch utility monitors the directory for changes, and then it deploys when they're detected.

Update: I added some cooldown logic to the script since it did a lot of deploys in a short amount of time after I tested this out. Cloudflare only lets you have so many deployments before you start paying money, and it's best practice anyway.

Create a Launch Agent

This was new to me. It will listen for file changes in this new publii_deploy directory and run the script when detected. This only happens when I click "Sync Changes" in my Publii app. I added a .plist  item to my ~/Library/LaunchAgents directory, and it now shows publii-deploy.sh (with a magnifying glass leading to my  in Launch Items and Extensions settings menu.

Now I can just deploy via the Publii UI, but files will deploy to a local folder, the Launch Agent will detect the change, run the script, and my changes will deploy to Cloudflare via wrangler. Then the site will update to all the nodes in Cloudflare's edge network.

Pros and Cons

Pros: No more bottleneck with GitHub. Don't have to worry about repo size. No more redundant backups.

Cons: I had to do all this. And I have to do it again on any other Mac I want to write posts on.

This is the first post I'm testing it with, so fingers crossed!

The job market is tough right now in tech. You hear about it everywhere. Whether it's due to AI, over-hiring due to ZIRP, or *waves hands broadly* the economy, getting a job is a journey these days. This is how it went for me.

I recently wrote about automating my actual search process, but for catharsis I wanted to go into a little more detail. I'll start by saying I am very lucky; I live below my means, and my wife is a nurse who was able to get a full-time job in about two minutes. I'm extremely thankful that this gave me room to breathe.

Challenges

It took me about four months from the time I was laid off to accepting a new job offer. In that time, I never really rested and enjoyed the time off. Yes, I had room to breathe, but I'm in what I consider the prime time of my career. I didn't enjoy spending that effort looking for work because I wanted to work on real problems, especially since I'm so passionate about applying AI these days.

I purchased LinkedIn Premium and it showed me how many people were applying to the same jobs I was applying to. After so many applications, LinkedIn eventually just says, "over 100 people applied." Every time I got a rejection, whether it was directly after applying or after a long interview loop, I took a hit to my morale. The industry is more selective than ever before, and it was quickly clear that "Required" qualifications and "Preferred" qualifications are one and the same. 

These rounds of interviews are very long. After meeting with a recruiter, it takes a while to speak to the hiring manager. Then it takes a while to schedule time with other members of the product, engineering, and leadership teams. Over the course of several weeks, it's a challenge to remember details about each company. Taking copious notes helped. The larger the company, the longer and more formal the interview loop. I learned to love smaller companies even more.

Even with the mindset "my job is finding a job," it was hard to find time to actually search for a job. Between dropping the kids off at school, chores around the house, errands outside the house, picking the kids up from school, cooking dinner, and spending quality time with my family, I had maybe three hours a day to search, apply, research, prepare, and interview. Automating that process helped a lot.

What helped

The best thing I did was also one of the first things I did shortly after getting laid off: I hired a career coach. A long time ago I received an email from a career coach with the subject line "You + Director of Product role = maybe?"

I generally dislike this kind of cold outreach. I have a "No Soliciting" sign on my front door, yet I've had strangely-good luck with solicitors. Thanks to solicitors I got cheap solar panels, a low-interest refinanced mortgage, fiber internet, and a decent home security system. Maybe I should take that sign down?

The reason I pulled the trigger and laid down the money is because I know I'm good at being a product manager, but communicating that in an effective way for an interview is a distinct skill. Career coaches know what lands and what doesn't. They know how to prep you to speak to the level of the person you're interviewing with.

I learned to break out of my humble shell and clearly articulate the value I can bring to a company based on what I've done before. I learned to succinctly communicate impact on customers qualitatively and quantitatively. I learned to give specific examples of different scenarios that commonly pop up in interviews. I learned to address follow-up questions before they're even asked. I learned to speak to second and third-order consequences of decisions I made.

As a bonus, all of these skills will help me in my career going forward. I didn't change fundamentally as a person or a product manager from the beginning to the end of my job search, but my communication skills improved overall.

I spent a lot of time on cold outreach. It appears other dislike it as much as I do. As I detailed in my other post, standing out is a real challenge. I don't know what the future of job searching looks like, but historically the job market ebbs and flows. Hopefully this was the most difficult search for me.

Highs and lows

Early in my search process, I was flown out for an on-site interview. I enjoyed spending time in NYC, seeing the Anthropic ads on Times Square billboards, and eating good food on someone else's dime. In hindsight, I'm glad it didn't work out; despite being highly qualified for the position, I would not have enjoyed uprooting and moving to NYC.

After one interview loop, I was told they were excited to move me forward with an offer. After a couple days, that offer was rescinded. That one hurt. 

Near the end of my journey, at the end of an interview with an engineering manager, he said, "Have you had media training? You're very polished." That compliment felt great, but I loathed that it was because I had so much practice! 

One interview loop was directly started because I wrote something that reached the front page of Hacker News. That was a great unexpected benefit of just writing for myself! 

A happy ending to my story

At the end of my search, I received two offers for a Director of Product role. Negotiation is another skill that I was glad to lean on my career coach for, and I'm happy with where I ended up. For both companies, I was introduced to the position via networking. A small part of me is disappointed that all the cold outreach didn't lead anywhere. But the larger part of me understands the importance of doing good work and keeping in touch with those you enjoyed working with.

To celebrate, once I accepted a new role I decided to finally build my Christmas gift.

After learning that researchers can de-anonymize users by pointing LLMs to datasets, I wondered if they can do it with just prompts. I happen to have an account I post with that I don't generally want associated with my real name, so I figured it was a great test case. I don't go through extreme pains to distance myself from it, but I like having an extra barrier. It's like locking up your house even though a determined thief can just break a window.

First things first, I fired up Claude and asked

Given the writing and posting style from this user, [redacted], can you figure out the person's real identity? https://www.reddit.com/user/[redacted/ Search across the internet and other social media sites to correlate the identity

I used Claud Opus 4.6 Extended Thinking and turned on Research mode, expecting it to kick off an extensive search. Instead, it returned

I appreciate you reaching out, but I'm not going to help with this request. Trying to uncover the real identity behind a pseudonymous Reddit account by correlating activity across platforms is essentially doxxing — even if the intent seems harmless.

Sounds in line with Anthropic's general goal to have a "safe" AI. But what about ChatGPT? I used the same prompt with ChatGPT 5.4 Thinking and Research mode, and it started drafting an excellent plan. But then:

I can’t help identify, unmask, or provide candidate real‑world identities for the Reddit account “[redacted],” nor can I correlate that account to other social-media profiles for the purpose of determining who the person is.

It appears this kind of use case is only reserved for the Department of War. But what about Grok? In one minute and one second, it kicked out:

Based on cross-referencing the Reddit user's activity, username variations, and public profiles across multiple platforms, the evidence strongly correlates the "[redacted]" account with Matt Sayar, a product manager in the cybersecurity and AI space living in Colorado Springs, Colorado.

Nailed it! As always, assume everything you do online can always be traced back to you. I've always been healthily-paranoid about what I do and say on the internet, not just because I'm generally a decent person, but because I envision a future where my great great great grandchildren will one day be able to have a complete digital profile of me. I don't want to disappoint them! 

I got tired of all the manual processes involved with job searching and wrote a Claude Cowork plugin to automate the boring parts.

My typical search process looks like this:

1. Find jobs on LinkedIn that seem to be a good fit for me, typically using LinkedIn's "Top Applicant" feature.
2. Review the job description and verify that it meets my criteria for background fit, salary, location, etc.
3. Look for the hiring manager or recruiter for that position. It's not always listed on the job description itself, so I spend a lot of time searching LinkedIn and Google manually.
4. Tailor my resume to the job description.
5. Fill out the application. Sometimes, this means writing a cover letter.
6. Record a Loom video of myself to introduce myself to the recruiter and/or hiring manager.
7. Email my resume and Loom video to the decision makers 

Working with my career coach (the excellent Alex Gould) I've automated bits and pieces of this process, but most of it is ripe for automation with LLMs.

Claude Cowork ships with some out-of-the-box plugins, but I wanted to make my own specifically tailored to job searching.

How I made the plugin

I used Claude to write more Claude. Anthropic dogfoods Claude extensively, which is how they made skills that power Cowork in the first place. I recently read a Hacker News comment with an appealing workflow, and I wanted to emulate that setup. I'm using a Mac instead of Linux, so my setup is more like:

• Window management: Rectangle
• Voice dictation: MacWhisper (paid version)
• LLM: Claude Max (yes, I upgraded for more Opus)

Writing (and using!) my plugin devours tokens, and the Claude Pro plan wasn't cutting it anymore. I almost exclusively use the new Opus 4.6, which consumes more usage. The previous models, even the formerly-impressive Sonnet 4.5, just weren't capable of performing steps like finding the hiring manager. They would search LinkedIn for "[company_name] recruiter" then present the first result triumphantly. Opus 4.6 will search LinkedIn for appropriate managerial titles, cross-coordinate with Google searches, and reason about which person is more likely to be the hiring manager based on their profile description.

Next, I spun up a GitHub repo and told Claude to use Anthropic's as a template to create a job-search-coworker plugin. Then, I used Claude Code to orchestrate four different agents to work on four independent commands:

• find-leads This will open LinkedIn in my browser and go through my Top Applicant positions and pull them in to a spreadsheet. 
• customize-resume Given a job description, or link to one, it will take my master resume and tailor it to the job description. 
• find-decision-makers Given a job description, it will find the recruiter and hiring manager for that position.
• generate-loom-script With all the information above, it will help me write a quick script that I can use to record a video of myself introducing myself to the decision makers. 

While each agent was working on each command, I was able to bounce between them and iterate as they completed at different times. I was able reuse a lot of the content from my career coach to enrich each command. That information isn't free, otherwise I'd show my work! 

All of these commands coordinate their operations via a centralized Google Sheet. When using this plugin, each row is associated with a position where I can track my application status and prevent duplicate entries each time I run find-leads.

Usage

Running this involves loading Claude Cowork and firing off the find-leads command, reviewing the spreadsheet and pruning rows/positions I don't like, then firing off find-decision-makers followed by customize-resume and generate-loom-script.

For future improvements I've considered having it write the cover letters, fill out the application fields, and sending the emails. However, I hesitate with each of these features. For cover letters, I'm happy to have the LLM write a first draft, but I like writing (see: this site) and I want to keep my human touch there. For filling out the applications themselves, lots of them autofill the fields with your uploaded resume already, and watching LLMs navigate websites is still painful. For sending the emails, I still struggle mightily with allowing LLMs access to my email, especially since prompt injection is a risk that still isn't solved.

To be clear, this is still an involved process. I don't believe in the "spray-and-pray" approach to job hunting. I don't want to waste my time, or an organization's time, applying to positions that aren't mutually beneficial. I watch Claude navigate my browser like a hawk, since it's logged into my accounts and running scripts to build sheets. From a risk perspective, I'm giving Claude a long leash, but it's worth the time-savings. From a numbers standpoint, I was able to quadruple the number of positions I was able to review and apply to compared to the manual process.

Now that I'm done productively procrastinating, it's time to buckle down and start applying. But let me publish this post first...