Step by step:

1. Turn off the bulb for 30s
2. Power cycle 6 times; on until it lights up, off immediately, count “one one thousand”, repeat
3. On the sixth on, leave it on and wait
4. When the bulb blinks it’ll be in pairing mode again

I didn’t have IPv6 enabled, and my ISP doesn’t assign IPv6, so I used unique-local-ipv6.com to get a unique prefix. Configured it as static in my local network setup, alongside the IPv4 configuration, on my UniFi Dream Machine and then all my bulbs connected.

Now my niece is learning to pronounce English colors as she cycles my apartment through every shade of the rainbow 😁

bd config set export.auto false

New projects get prompted during init, only existing projects will get this nice surprise.

If you’re using dolt’s git export instead, commit the config change and delete the issues.jsonl file.

I found this error message when starting dolt up after upgrading the server: invalid journal record at offset 5831169: invalid journal record: CRC checksum does not match.

To recover, go to your dolt DB:

cd ~/.beads/shared-server/dolt/<db> \
  && dolt fsck --revive-journal-with-data-loss

It backs up the journal, truncates to the point it failed, and makes it possible to start the server again. In my case the db came back up with data loss. I did a bd dolt pull and it pulled in what I had pushed most recently, so nothing major was lost.

I’m guessing this happened because installing beads 1.0 also upgraded dolt. If you’re upgrading, it’s worth making sure you’ve done bd vc commit && bd dolt push in all repos beforehand. That means configuring a git remote first.

You’ll get Error 1105 (HY000): cannot merge with uncommitted changes and it’s confusing when the beads project you’re working in has autocommit on. The uncommitted changes are in a different project that’s also using the shared server.

The fix is to go through your other projects and commit their changes (or enable autocommit by adding dolt.auto-commit: "on" to .beads/config.yml). To find which databases have uncommitted changes:

query=$(bd sql "SHOW DATABASES" 2>/dev/null \
  | grep -v -E "^(Database|--|\(|dolt|information_schema|mysql)" \
  | while read db; do
      echo "SELECT '$db' as db, table_name, status FROM \`$db\`.dolt_status WHERE table_name != 'config'"
    done \
  | paste -sd'|' - | sed 's/|/ UNION ALL /g')
bd sql "$query"

For each project that shows up, go in and run bd vc commit -m 'Snapshot', and then you can pull again.

We’re ignoring changes to the config table because they don’t seem to matter for pulling.

1. bd dolt stop: stop the server in the git repo you’re restoring so you can operate on its database
2. cd ~/.beads/shared-server/dolt: go to the shared dolt server (it’s .beads/dolt/beads otherwise, I think)
3. rm -rf <prefix>: remove the DB if it exists already
4. dolt clone git+ssh://git@github.com/user/repo.git <prefix>: clone it to your prefix
5. cd - && bd ready: see if you can see the listing

I don’t know why my local just refused with bd bootstrap but this allowed me to fix it manually.

I was running in “a server per repo” mode and it’s just unnecessary, they clash in annoying ways when switching repos, and once I understood how beads store data in git I embraced dolt and the shared server. I had hesitated because I had somehow connected it with DoltHub, and that’s optional for using beads.

1. bd backup: make sure you have your local database available for restore
2. bd dolt stop: so we don’t conflict on anything
3. bd dolt set shared-server true: to configure this project to use the shared server
4. bd restore: to restore the db

For most of my projects this just worked, but in some cases I had to reset the dolt configuration for the repo.

After bd init, your dolt data only exists locally. You can check with bd dolt remote list which will show “No remotes configured.”

Point it at your existing git repo:

bd dolt remote add origin git+ssh://git@github.com/you/repo.git

Then push and verify:

bd dolt push
git ls-remote origin 'refs/dolt/*'
# 52689c0ae7...  refs/dolt/data

I’ve seen refs/ used for GitHub PRs before (refs/pr/<num>/head) but I haven’t looked into how any of it actually works. For now, seeing refs/dolt/data show up in git ls-remote is enough to make me comfortable that my data exists somewhere.

To avoid doing this manually on every machine, add the remote to .beads/config.yaml:

sync:
  git-remote: git+ssh://git@github.com/you/repo.git

Anyone cloning the repo can then run bd bootstrap and bd dolt push / pull will just work.

As of beads 0.62.0, this is what I run:

# configure agents later
bd init --shared-server --skip-agents

# setup a CLAUDE.md and configure claude hooks to re-prime beads
bd setup claude --project
bd hooks install

# move local settings to project settings
mv .claude/settings{.local,}.json

# amend init commit so .beads-credential-key never enters git
git add .claude/settings.json CLAUDE.md
git commit --amend -CHEAD

# share beads data alongside your git repo
bd dolt remote add origin git+ssh://git@github.com/you/repo.git
echo -e 'sync:\n  git-remote: git+ssh://git@github.com/you/repo.git' >> .beads/config.yaml

The key thing is --skip-agents and then bd setup claude --project instead, so you get Claude-specific hooks rather than the generic agent integration. The dolt remote lines are so your beads data gets stored alongside your git repo.

The errors look like this:

Error: failed to open Dolt store: database "bs" not available after CREATE DATABASE: Error 1049 (HY000): database not found: bs

or:

Error: failed to open database: database "bs" not found on Dolt server at 127.0.0.1:3308

Fix it by stopping dolt and removing its local state:

bd dolt stop  # or you won't be able to delete all the files
rm -rf .beads/dolt/
rm .beads/dolt-server.*

Also check .beads/metadata.json for stale server connection config. Mine had extra fields that were messing things up. I cleaned it down to just the essentials:

{
  "database": "dolt",
  "backend": "dolt",
  "dolt_mode": "server",
  "dolt_database": "bs",
  "project_id": "deadbeef-9164-426d-8c1c-ff8ae5d35e3e"
}

Then reinit: bd init --force --prefix <yourprefix>.

bd doctor --fix couldn’t sort this out, which is why I ended up doing all this manual cleanup.

I hit this upgrading to beads 0.61.0 when switching to shared-server mode.

• Top: the first print, fully believing I had nailed everything on the first try (also some hope, it was printing over night so hey, it might mean I’m done with this first design)
• Middle: second print, cut smaller but enough to let the device rest. Made me realize I hadn’t taken into account that the device tapers at the top and bottom, so it needs less living space
• Bottom: latest print, cut to print as small as possible, and maybe too small, but it seems to indicate this is good. Next up, a full print

Next time, reverse the order 😅

The way I have historically managed the hooks is by having a script/bootstrap that symlinks scripts, and then hopefully my team mates run bootstrap when something changes.

So this is a nice way of making that entire problem go away.

I had done a server migration and completely missed the .env file when copying, and after recreating it all started working.

{
  "permissions": {
    "allow": [
      "Bash(pnpm test:unit:*)",
      "Bash(pnpm test:e2e:*)"
    ]
  },
  "sandbox": {
    "excludedCommands": [
      "pnpm test:unit:*",    // Same commands, different purpose
      "pnpm test:e2e:*"
    ]
  }
}

Note: Tested with Claude Code v2.1.3

For example:

{
  "permissions": {
    "allow": [
      "Bash(pnpm test:unit:*)", // Allows pnpm test:unit --arg etc.
      "Bash(pnpm test:e2e:*)",
      "Bash(pnpm test*)"        // ❌ <- will allow testAndDestroyEnvironment
    ]
  }
}

Note: Tested with Claude Code v2.1.3

To make linking internally easier, I created an Alfred app snippet ({{< relref “{clipboard}” >}}), so now I only have to open the other post, double-tap shift, type “source root”, then run the snippet, and all linked up.

With the expression //*[@class="published-at"] you will get the HTML tag that has class="published-at", and if you then want the datetime property you add /@datetime to get it.

I realized this while trying to calculate what height TV bench I should get:

Annoyingly, you can’t declare a variable and see its result in one step. I found two options for this, and you decide which you prefer.

If you’re okay with an extra line you can show it by declaring the variable again, just let notes fill in the part after the =:

MyValue=25*4
MyValue=100

Or, instead, write the calculation first, and let Notes show the result, then add the variable name.

First, do the calculation:

25*4=<wait for Notes to autofill>

Second, name the variable by adding the name and = at the front of the line:

MyValue=25*4=100

I naturally ended up doing this second while I was working it through, since I wanted to see the result first, and then I named the variable.

Notes also handle some unit conversions for you:

The average daily temperature in Singapore as Fahrenheit is: 
0f + 32c = 89.6f

How many Swedish Krona do I get from one SGD right now?
0sek + 1sgd = 7.22 SEK

I started on my phone taking Notes, and was surprised to realize I didn’t have to fire up Python on my computer to finish it. A new useful tool that I always carry with me. 🙂

Documentation links to Apple:

• for iOS
• for macOS

So the below won’t match because it’s looking for a literal $ in the string, instead of matching at the end of the string:

file="https://example.com/blog-post/index.html"
[[ "$file" =~ "/index.html$" ]] && echo MATCH

But this will:

file="https://example.com/blog-post/index.html"
[[ "$file" =~ /index\.html$ ]] && echo MATCH

Note: [[ ]] is a Bash extension of the test command that supports regexes (among other things). The [ ] is the POSIX standard test (see man test), so [[ ]] features may not work in other shells or in limited shells like ash or sh.

I learned this while removing index.html from file paths to bust Cloudflare’s cache for directory listings/pretty URLs. I then used parameter expansion to replace the match with an empty string, so directory URLs are also purged:

echo ${file/index.html/}

I used to think there was some magic rule about how | decided where to split, but it’s simply: characters and subpatterns concatenate into a single pattern first, then | splits the entire thing left and right unless you explicitly group it. That’s because | has the lowest precedence so all other operations happen first.

So while I used to think that prefix_cat|dog would match prefix_cat and prefix_dog, it actually matches prefix_cat and then dog.

And that exact problem in a real-world example, to find if a page is linking to another, both relative or full URL:

The regex <a[^>]+href=["']https://www.example.com/page|/page will match /page, just as in the dog example, it’s splitting the whole pattern and isn’t what I want.

See it in action with the excellent Regex101.com:

To make it do what I want I need to constrain it in a capturing group (pattern), and since I don’t want to reference the group later, I constrain it inside a non-capturing group (?:pattern).

That gives me the pattern <a[^>]+href=["'](?:https://www\.example\.com/page|/page), and then it matches both the full URL (https://www.example.com/page) and the relative URL (/page):

I love Regex101.com. 🙂 I have commented gnarly regexes in my code with links to it, so colleagues can paste them there and understand what’s going on. The sidebar explains what each subpattern does, and the color highlighting shows what matches what in real-time.

Example: When modifying Mastodon/Bluesky cross-posting nodes that depend on an RSS Feed Trigger, execute and pin the RSS node. Then iterate on the posting logic that only depends on the RSS data without triggering any posting.

Where my confusion came from: I haven’t figured out fan-out and fan-in, because I would expect to run my cross-posting in parallel and not sequentially, but… I have seen there is some kind of fan-out, but how to use it? 🤷

So my assumption that I wouldn’t get the data reloaded if I only ran the first step was incorrect, and I’ve been a bit hesitant to test since there’s only production (this personal stuff doesn’t have a proper testing env 😅).

I kept my assets/resources in static/ and didn’t get why I always had to write /absolute/urls and use hope as a strategy, when I moved them into assets/ I could look up all my assets.

The below would fail because it couldn’t find the asset:

{{ with resources.Get "/author/bjorn.jpg" }}
  <img src="{{ .RelPermalink }}" alt="Björn Andersson">
{{ end }}

Hugo has global assets accessed through resources.Get which only looks up in assets/. Then page assets that you get through .Page.Resources.Get instead, which I haven’t used yet.

If you want to use both, then do like the figure shortcode and do it in an or:

{{- with or (.Page.Resources.Get $u.Path) 
            (resources.Get $u.Path) -}}
{{- $src = .RelPermalink -}}
{{- end -}}

Then groupmod -n newgroup oldgroup if you need to rename the group.

I had a vague memory of being able to change settings (shell etc.) by updating using vipw so I renamed there and then edited /etc/groups but I then got an error when trying to become that user:

# su bjornnow
su: Authentication failure

To fix it, I had to update all these files (and rename the home folder):

• vipw: updates /etc/passwd
• vipw -s: updates /etc/shadow which contains the password hashes and are only readable by root
• vigr: updates /etc/group
• vigr -s: updates /etc/gshadow the same as passwd (also TIL, groups can have passwords)

This is what you get when you think you remember how things work, and didn’t want to look up how usermod worked, a little bit of knowledge is dangerous and all… 😅

This seems to be a security decision to avoid leaking which certificates (and therefore sites) it hosts by not providing a default (can’t find a source and it’s what the search previews give me). Which my old nginx setup did.

The cryptic error message from cURL, which I didn’t find anything obvious online about:

curl: (35) OpenSSL/3.0.13: error:0A000438:SSL routines::tlsv1 alert internal error

And I debugged this together with Claude by enabling debug logging in Caddy ({ debug } at the top level in the Caddyfile) which gave me this output when I called it using cURL:

2025/10/01 10:04:56.330 DEBUG   events  event   {"name": "tls_get_certificate", "id": "7353980b-a85e-4f10-9704-89da7358ef89", "origin": "tls", "data": {"client_hello":{"CipherSuites":[4866,4867,4865,49196,49200,159,52393,52392,52394,49195,49199,158,49188,49192,107,49187,49191,103,49162,49172,57,49161,49171,51,157,156,61,60,53,47,255],"ServerName":"","SupportedCurves":[29,23,30,25,24,256,257,258,259,260],"SupportedPoints":"AAEC","SignatureSchemes":[1027,1283,1539,2055,2056,2057,2058,2059,2052,2053,2054,1025,1281,1537,771,769,770,1026,1282,1538],"SupportedProtos":null,"SupportedVersions":[772,771],"Conn":{}}}}
2025/10/01 10:04:56.330 DEBUG   tls.handshake   no matching certificates and no custom selection logic  {"identifier": "127.0.0.1"}
2025/10/01 10:04:56.331 DEBUG   tls.handshake   all external certificate managers yielded no certificates and no errors {"remote_ip": "127.0.0.1", "remote_port": "42516", "sni": ""}
2025/10/01 10:04:56.331 DEBUG   tls.handshake   no certificate matching TLS ClientHello {"remote_ip": "127.0.0.1", "remote_port": "42516", "server_name": "", "remote": "127.0.0.1:42516", "identifier": "127.0.0.1", "cipher_suites": [4866, 4867, 4865, 49196, 49200, 159, 52393, 52392, 52394, 49195, 49199, 158, 49188, 49192, 107, 49187, 49191, 103, 49162, 49172, 57, 49161, 49171, 51, 157, 156, 61, 60, 53, 47, 255], "cert_cache_fill": 0.0002, "load_if_necessary": true, "obtain_if_necessary": true, "on_demand": false}
2025/10/01 10:04:56.332 DEBUG   http.stdlib     http: TLS handshake error from 127.0.0.1:42516: no certificate available for '127.0.0.1'

The culprit was the "ServerName":"", and I’m guessing also "sni": "", so if you need to validate and see that as well then that’s likely the problem. 🙂

This jogs my memory, back in the olden times, we would need a unique IP address for each HTTPS site because we couldn’t multi-host, and SNI solved that problem.

Importantly, SNI doesn’t encrypt the host name and it’s still sent in plaintext to the server.

:focus-visible only shows focus styling when the browser determines keyboard navigation is being used, unlike :focus which triggers on all interactions including mouse clicks.

button:focus-visible { outline: 2px solid blue; }

This prevents the “clicked button stays highlighted” problem while preserving accessibility for keyboard users. Full browser logic on MDN.

This seems to be based on EU law, which makes sense, so I’ll assume the gist of this is true in other EU countries too.

This has to do with “producentansvar” (“producer responsibility”) which by law requires producers of products to be responsible for the recycling of the packaging, by joining a “collection system that takes operative and financial responsibility for collecting and managing packaging material."1 § 3

This sounds like systems thinking in play: By making producers responsible for recycling, they’re incentivized to create packaging that’s both easy to identify/recycle (cheaper for them) and easy for consumers to sort correctly (to hit reuse targets).

I watched Chris Spargo’s video on Why recycling in England is so complicated? yesterday, which explains how England is going through this shift to Extended Producer Responsibility (EPR) right now, and without seeing that video it wouldn’t have clicked as easy when I was reading this now.

(The egg shells don’t qualify as food waste while bones do, I don’t know what to make of that, and apparently it’s the opposite in my mom’s municipality 🤷)

Their demo:

For example, if you run:

gum input --width 50 --header "Favorite author?" \
 --value "Terry Pratchett"

It will give you a question with a sensible default that you can then change as you want, and what you wrote as you hit enter will be returned so your script can then use it.

If you instead has some options where you want to select one:

gum choose --header "Favorite book?" \
 "Hogfather" "Thief of Time" "The Night Watch" "Small Gods"

Which gives you a selection box, you can configure to allow multiple selections, and the one you pick is then returned.

And one I really love, have it show a spinner to indicate that yes… something is still going on:

gum spin --title="Counting down from 10" sleep 10

Built on bubbletea, a Go TUI framework, use that directly if you need these components in Go code rather than shell scripts.

When bash is started non-interactively, to run a shell script, for example, it looks for the variable BASH_ENV in the environment, expands its value if it appears there, and uses the expanded value as the name of a file to read and execute. Bash behaves as if the following command were executed:

if [ -n "$BASH_ENV" ]; then . "$BASH_ENV"; fi

but the value of the PATH variable is not used to search for the file name.

world: &world World
example: &example-anchor
  HELLO: *world
  There: Yo

my-values:
  <<: *example-anchor
  foo: bar

becomes

---
example:
  HELLO: World
  There: Yo
my-values:
  HELLO: World
  There: Yo
  foo: bar

So if you have a document like below, there is no syntax to make the commands a three item list:

---
example: &example
  - "Hello"
  - "World!"

name: "hello"
commands:
  - << *example
  - "Oho!"

WILL NOT turn into:

---
example: &example
  - "Hello"
  - "World!"

name: "hello"
commands:
  - "Hello"
  - "World!"
  - "Oho!"

…and just because before I found the issue where it was described that this isn’t happening I had made a test repo to try and understand if it was a library/usage issue, because I noticed I could get things running on CI with << *example which syntax didn’t give syntax error, but also did nothing when running on Drone CI.

Just put your shebang as /usr/bin/env uv run python3 and it always runs in the virtualenv, no wrapper script needed. This feels obvious in hindsight, it’s what you expect from these tools 😃

This is different from normal shell where ${VAR} is the same as $VAR and it exists to make sure you can concatenate strings without issues, i.e.

VAR=m000
echo "'${VAR}est'"  # => 'm000est'
echo "'$VARest'"    # => ''

I.e., with "$@" the arguments "one two" three will be 2 arguments, the first being "one two", and without it will become three arguments, all separated by space.

Go to Settings -> Webhooks and then click on the integration you want to look at.

1. cp /etc/pam.d/sudo_local{.template,}
2. Edit /etc/pam.d/sudo_local and uncomment auth sufficient pam_tid.so.

The reason for doing this in sudo_local is that this file will not get reset with system changes from Apple.