Serenity on ¬ just serendipity 🍀

★ logseq → obsidian

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 20 Jul 2026 18:26:33 +0200

Problem statement: LogSeq v2 is moving its graph format from markdown files to a sqlite database:

This is not OK for my workflow.

The whole point of adopting LogSeq was, just like Obsidian, the promise of managing my notes in plain markdown files.

Even though this migration is supposed to improve the performance of the app, it ends up breaking a fundamental reason why I decided to use it in the first place.

Furthermore, the Second Brain app of my choice is supposed to be boring. It is a big deal to wake up one day and find that I can no longer open my notes graph with the new version of the app. This is a no-go.

Well, classic LogSeq (dubbed “Logseq OG”) will be kept around, but I can only assume that from this point on it will become poorly maintained — it will never get the same amount of resources and attention as the new version.

Also, even if I really wanted to use the new version, it turns out it does not work on mobile (iOS), which is a dealbreaker, therefore I would be forced to stay stuck with OG for a while. Heck, the iOS app hasn’t been updated in 2y anyway, and it shows.

There will be no hard feelings, this is a peaceful breakup; I enjoyed LogSeq, it’s simpler and more streamlined than Obsidian; as such, I will keep recommending it to most people.

That said, I’ve always been pondering whether to switch back to Obsidian. There’s no doubt that Obsidian is more actively maintained and better suited for Gen AI / LLM workflows these days (see Obsidian CLI for starters).

Why did I switch to LogSeq from Obsidian in the first place? Because Obsidian was too much for me. It was more complicated (in terms of its feature set) than warranted for my purposes. It offered more than I needed. That’s not necessarily a bad thing. It’s like saying (hypothetically) “I prefer nano instead of vim”, or that I prefer to live in a studio apartment than in a 3-bedroom house.

Obsidian won’t let me down in terms of changing its file format: file over app is one of their core promises:

File over app is a philosophy: if you want to create digital artifacts that last, they must be files you can control, in formats that are easy to retrieve and read. Use tools that give you this freedom.

File over app is an appeal to tool makers: accept that all software is ephemeral, and give people ownership over their data.

This is very important. My Second Brain app must not lock me in. LogSeq, upon the switch to a sqlite database, breaks this invariant. I want — I need — boring markdown files. Do not break the beauty and simplicity of this setup.

Migration #

LogSeq and Obsidian are similar but not identical.

In order to bridge the gap, a small python script (drafted by Claude Fable 5, reviewed by me) walked the existing LogSeq Graph and wrote to a fresh Obsidian Vault, leaving the source untouched, should I decide to roll back.

These are some mappings (LogSeq → Obsidian):

journals/2024_03_04.md         ->  journals/2024-03-04.md
pages/Foo___Bar.md             ->  Foo/Bar.md          (namespaces)
- TODO buy milk                ->  - [ ] buy milk
- DOING ...                    ->  - [/] ...
- DONE ...                     ->  - [x] ...
[[Feb 9th, 2026]]              ->  [[2026-02-09]]
((67ea7dba-3200-...))          ->  [[2025-03-31#^67ea7dba]]
{{embed ((67ea7dba-3200-...))}} -> ![[2025-03-31#^67ea7dba]]
{{video https://...}}          ->  plain URL
id:: 67ea7dba-3200-...         ->  ^67ea7dba block anchor (when referenced)
collapsed:: true, :LOGBOOK:    ->  dropped

Plus one line of config so that the LogSeq journals become Obsidian daily notes:

{
    "folder": "journals",
    "format": "YYYY-MM-DD"
}

…and then I spent some time trying to dumb down Obsidian, disabling core plug-ins I do not need at this time. There’s no need for community plug-ins for now (if ever).

So far, I only migrated my corp (work) notes. I’ll need to migrate my personal PKM as well, but I am happy to simply pilot this migration at work first, so that I don’t get to regret twice.

The migration script:

#!/usr/bin/env python3
"""Migrate LogSeq graph -> Obsidian vault. Source is never modified."""
import re
import shutil
from pathlib import Path

SRC = Path("/Users/thiago.perrotta/Library/Mobile Documents/iCloud~com~logseq~logseq/{redacted}")
DST = Path("/Users/thiago.perrotta/Library/Mobile Documents/iCloud~md~obsidian/{redacted}")

UUID = r"[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}"
MONTHS = {m: i + 1 for i, m in enumerate(
    ["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"])}

MARKER_MAP = {
    "TODO": "[ ]", "LATER": "[ ]", "WAITING": "[ ]",
    "DOING": "[/]", "NOW": "[/]", "IN-PROGRESS": "[/]",
    "DONE": "[x]",
    "CANCELED": "[-]", "CANCELLED": "[-]",
}


def journal_name(stem: str) -> str:
    return stem.replace("_", "-")


def dest_rel(src_file: Path) -> Path:
    """Map a source md file to its vault-relative destination path."""
    if src_file.parent.name == "journals":
        return Path("journals") / (journal_name(src_file.stem) + ".md")
    # namespace pages: A___B.md -> A/B.md; everything else under pages/
    parts = src_file.stem.split("___")
    return Path(*parts[:-1]) / (parts[-1] + ".md") if len(parts) > 1 else Path("pages") / src_file.name


def page_ref(src_file: Path) -> str:
    """Wiki-link target for a source file (what goes inside [[...]])."""
    if src_file.parent.name == "journals":
        return journal_name(src_file.stem)
    return "/".join(src_file.stem.split("___"))


def collect_ids():
    """uuid -> (link target, anchor) for every id:: definition; also which uuids are referenced."""
    defined, referenced = {}, set()
    for f in sorted(SRC.glob("pages/*.md")) + sorted(SRC.glob("journals/*.md")):
        text = f.read_text(encoding="utf-8")
        for m in re.finditer(rf"^\s*id:: ({UUID})\s*$", text, re.M):
            defined[m.group(1)] = (page_ref(f), m.group(1)[:8])
        for m in re.finditer(rf"\(\(({UUID})\)\)", text):
            referenced.add(m.group(1))
    return defined, referenced


def convert(text: str, defined, referenced) -> str:
    lines = text.split("\n")
    out = []
    in_fence = False
    in_logbook = False

    def resolve_ref(m):
        uuid = m.group(1)
        if uuid in defined:
            target, anchor = defined[uuid]
            return f"[[{target}#^{anchor}]]"
        return m.group(0)

    def resolve_embed(m):
        uuid = m.group(1)
        if uuid in defined:
            target, anchor = defined[uuid]
            return f"![[{target}#^{anchor}]]"
        return m.group(0)

    def date_link(m):
        mon, day, year = m.group(1), int(m.group(2)), int(m.group(4))
        return f"[[{year:04d}-{MONTHS[mon]:02d}-{day:02d}]]"

    pending_merge_indent = None  # indent of an empty bullet awaiting its continuation line

    for line in lines:
        stripped = line.strip()

        # merge continuation text into a bullet whose first line was a property
        if pending_merge_indent is not None:
            m = re.match(rf"^{re.escape(pending_merge_indent)}\s\s?(\S.*)$", line)
            pending_merge_indent = None
            if m and out:
                out[-1] = out[-1] + m.group(1)
                continue

        # LogSeq opens fences on bullet lines ("- ```lang"), closes with bare "```"
        if re.match(r"^\s*(?:[-*+]\s+)?```", line):
            in_fence = not in_fence
            out.append(line)
            continue
        if in_fence:
            out.append(line)
            continue

        # LOGBOOK drawers
        if stripped == ":LOGBOOK:":
            in_logbook = True
            continue
        if in_logbook:
            if stripped == ":END:":
                in_logbook = False
            continue

        # bullet-form property lines: block whose first line is the property itself
        # ("- collapsed:: true" with real content on the continuation line below)
        m = re.match(rf"^(\s*)- (collapsed:: (?:true|false)|logseq\.order-list-type:: number|id:: ({UUID}))\s*$", line)
        if m:
            indent = m.group(1)
            uuid = m.group(3)
            if uuid and uuid in referenced:
                out.append(f"{indent}- ^{uuid[:8]}")
            else:
                out.append(f"{indent}- ")
                pending_merge_indent = indent
            continue

        # property lines
        m = re.match(rf"^\s*id:: ({UUID})\s*$", line)
        if m:
            uuid = m.group(1)
            if uuid in referenced and out:
                # anchor goes at end of the block's first line
                for i in range(len(out) - 1, -1, -1):
                    if out[i].strip():
                        out[i] = out[i].rstrip() + f" ^{uuid[:8]}"
                        break
            continue
        if re.match(r"^\s*collapsed:: (true|false)\s*$", line):
            continue
        if re.match(r"^\s*logseq\.order-list-type:: number\s*$", line):
            # convert the previous bullet line to a numbered item; number it by
            # walking back past deeper lines to the previous same-indent sibling
            for i in range(len(out) - 1, -1, -1):
                bm = re.match(r"^(\s*)- (.*)$", out[i])
                if bm:
                    indent = bm.group(1)
                    n = 1
                    for j in range(i - 1, -1, -1):
                        if not out[j].strip():
                            continue
                        lead = out[j][: len(out[j]) - len(out[j].lstrip())]
                        if len(lead) > len(indent):
                            continue  # child or continuation line
                        if lead == indent:
                            nm = re.match(rf"^{re.escape(indent)}(\d+)\. ", out[j])
                            n = int(nm.group(1)) + 1 if nm else 1
                        break
                    out[i] = f"{indent}{n}. {bm.group(2)}"
                    break
            continue

        # task markers on bullet lines (not headings)
        line = re.sub(
            r"^(\s*)- (TODO|LATER|WAITING|DOING|NOW|IN-PROGRESS|DONE|CANCELED|CANCELLED)\s+",
            lambda m: f"{m.group(1)}- {MARKER_MAP[m.group(2)]} ",
            line,
        )

        # drop LogSeq priority markers
        line = re.sub(r"\[#[A-C]\]\s*", "", line)

        # embeds and block refs
        line = re.sub(rf"\{{\{{embed \(\(({UUID})\)\)\}}\}}", resolve_embed, line)
        line = re.sub(r"\{\{embed \[\[([^]]+)\]\]\}\}", r"![[\1]]", line)
        line = re.sub(rf"\(\(({UUID})\)\)", resolve_ref, line)
        line = re.sub(r"\{\{video ([^}]+)\}\}", r"\1", line)

        # date links: [[Feb 9th, 2026]] -> [[2026-02-09]]
        line = re.sub(
            r"\[\[(Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec) (\d{1,2})(st|nd|rd|th), (\d{4})\]\]",
            date_link, line)

        # asset paths
        line = line.replace("](../assets/", "](assets/")

        out.append(line)

    result = "\n".join(out)
    if result.strip() == "-":  # LogSeq empty-page placeholder
        result = ""
    return result


def main():
    defined, referenced = collect_ids()
    orphans = referenced - set(defined)
    if orphans:
        print(f"WARNING: refs without target: {orphans}")

    written = 0
    for f in sorted(SRC.glob("pages/*.md")) + sorted(SRC.glob("journals/*.md")):
        rel = dest_rel(f)
        dst = DST / rel
        dst.parent.mkdir(parents=True, exist_ok=True)
        dst.write_text(convert(f.read_text(encoding="utf-8"), defined, referenced),
                       encoding="utf-8")
        written += 1

    assets_dst = DST / "assets"
    assets_dst.mkdir(exist_ok=True)
    copied = 0
    for a in SRC.glob("assets/*"):
        if a.name == ".DS_Store":
            continue
        shutil.copy2(a, assets_dst / a.name)
        copied += 1

    print(f"wrote {written} notes, copied {copied} assets")


if __name__ == "__main__":
    main()

— § —

Reply via email

#ai #bestof #dev #pkm #serenity

Sports.app

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 20 Jul 2026 11:35:19 +0200

♠ The recently released Apple Sports app was surely a great way to keep up with the World Cup on the go.

It is very polished; it does one thing, and does it well. Simple yet effective. I do recommend it.

It’s especially handy when you’re not able to watch a match live for whatever reason.

If you’re able to watch a match in real time then it’s better to leave your phone in airplane mode, otherwise chances are you’ll get a goal notification from the app before you get to see it on the TV due to broadcasting delays. Wild.

— § —

Reply via email

#serenity

★ The lifecycle of physical books

serendipity@perrotta.dev (Thiago Perrotta) — Sat, 18 Jul 2026 14:21:02 +0200

♠ Typically:

Step 1: buy a physical book — via Amazon, bookstores, classifieds marketplaces (e.g. Craigslist in the US, Kijiji in Canada, Kleinanzeigen in Germany). Bonus points for used copies. This should take up to a week.
Step 2: read the ~~goddamn~~ book. This is arguably the hardest part, albeit the most exciting one. As Alex Kotliarskyi puts it (emphasis mine):

The 80/20 rule says that 80% value come from 20% effort. What if we apply the 80/20 rule to the first 80%?

So 80% of 80% of value come from 20% of 20% of effort. Or 64% of the value come from 4% of the effort. Apply it again and you’ll get ~51% value from ~1% effort. Which is ballpark close to “Dimidium facti, qui coepit, habet” — Well begun is half done.

Once you start, it becomes much easier to go through it, I promise.

Step 3: Find a way to get rid of the book.

There’s no shame. Have gratitude, appreciate it, and then simply move on. I don’t like to stick with physical books forever. They unnecessarily occupy space in my home.

There are various ways to do so:

Donating is the easiest. Leave books by the vestibule of your apartment building complex with a post-it note stating that it’s free¹.
Leave books in public community bookshelves or libraries.
Gift a book to a friend. Bonus points for being thoughtful.
Sell a book in a classified site. Bonus points for using the same one you got it from.

And now, adding the missing alternative to my repertoire:

momox:

Schnell & einfach Gebrauchtes verkaufen Was möchtest du verkaufen? Bücher & Medien Kleidung

Momox is

a Berlin-based recommerce site for used books and media founded by Christian Wegner in 2004.

In plain English: it’s like selling a book in a classified, but without having to wait for someone to actually message you².

Scan or input its ISBN, get an estimate of how much you’ll get for it (assuming it’s in good condition), arrange a box, pack the book, ship it for free, wait, and then get paid. It’s very streamlined and easy to use.

So far, I’ve only heard of this concept in Germany. I would love to find out if it’s possible to follow suit in other countries.

Zu verschenken, as they say it auf Deutsch. In plain English, “for free” suffices. ↩︎
And eventually lowball or flake on you. ↩︎

— § —

Reply via email

#bestof #serenity

★ Project Aces OST

serendipity@perrotta.dev (Thiago Perrotta) — Fri, 17 Jul 2026 12:52:54 +0200

♠ Project Aces (from the Ace Combat series) has one of the best videogame OSTs ever.

I am¹ finally pleased² to find their whole collection available on Spotify as of 2026. Previously, only a handful of them were available.

No more fiddling random playlists on YouTube Music!

↗ Gerrymandle: daily gerrymandering puzzle

serendipity@perrotta.dev (Thiago Perrotta) — Fri, 26 Jun 2026 11:30:06 +0200

Welcome to Gerrymandle!

Draw electoral district lines to win more seats than your opponents and win the election!

Win the election for purple. Draw 5 districts of 6 populated tiles so that they win more districts than any other party.

This is fun! And eye-opening.

— § —

Reply via email

#gaming #serenity %commentary

Stop Energy

serendipity@perrotta.dev (Thiago Perrotta) — Sun, 14 Jun 2026 17:41:56 +0200

♠ RFC: What is Stop Energy? by Dave Winer:

Suppose someone, call him Mr. A, has an idea that he believes is ready to deploy, or is requesting comments as he is getting ready to deploy. So he posts an RFC, usually on a mail list or a website, in the hope that people will spot a problem and help him figure out a solution; or find no problems and co-develop an implementation, or develop a compatible implementation. In theory, the Internet is a collegial environment, with lots of people who want to do new stuff, where one should expect to get this kind of help.

In this scenario, A is a proponent of Forward Motion. In all likelihood, instead of getting help, A will encounter Stop Energy, reasons why he can’t or shouldn’t be allowed to do what he proposes.

Stop Energy is not reasoned, it never takes into account the big picture, it is the mirror image of Forward Motion. In the Stop Energy model, everyone, no matter how small their stake in a technology, has the power to veto. Nothing ever gets done, and people who want to move forward are frustrated in every attempt to move. Unfortunately, Stop Energy is the rule, not the exception.

— § —

Reply via email

#dev #serenity

Don't ask to ask

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 10 Jun 2026 12:02:06 +0200

♠ Just ask! Just do it™:

[…]

So, to summarize, don’t ask “Any Java experts around?”, but rather ask “How do I do {problem} with Java and {other relevant info}?”

Previously.

— § —

Reply via email

#serenity

terraform police

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 08 Jun 2026 01:21:44 +0200

♠ The Master wakes to a pager. Coffee is ready.

Apprentice: “Master, who created the drift?”

Master: “Which drift?”

Apprentice: “The bucket tag — live state ≠ terraform config in git.”

Master: “Show me the chain.”

They follow: pipeline → cronjob → github actions container → terraform CI bot → github pull request → human.

Apprentice: “We found them. Shall we ~~retaliate~~ punish?”

Master: “What would punishing do to the state?”

Apprentice: “Teach them not to drift.”

Master: “Teach the pipeline to ask ‘why’ first.”

Apprentice: “And what if entropy persists?”

Master: “Record why. The record is a lamp in the fog.”

Apprentice: “Who made the first drift in all systems?”

Master: “Entropy.”

Apprentice: “Then we’ll blame entropy?”

Master: “Blame nothing. Write the JIRA ticket.”

The apprentice writes the ticket. The pipeline asks “why” and then sleeps. The pager blinks once. Dawn sets. Just another ordinary day goes by.

What does the “Terraform police” do, you asked?

It pursues active drifts in the terraform repository.

It chases the individual(s?) responsible for deepening chaos.

It teaches them how not to do so anymore.

Rinse and repeat.

Will the workload of the “Terraform police” ever end?

Is drift an ephemeral state?

Is this world a simulation?

Is this a world simulation?

— § —

Reply via email

#dev #serenity #terraform

Gmail: disable AI inbox

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 27 May 2026 14:03:16 +0200

♠ Google: First, it’s Gen AI, not “““AI”””.

You’ve already been doing machine learning in our inboxes for ages now.

Second: no, thank you.

Settings > Inbox > AI Inbox > [ ] Show AI Inbox:

Show AI Inbox — See a separate view of your inbox that includes suggested to-dos and key updates from your emails, chats, documents and more.

Uncheck that checkbox, and then continue to live happily and free¹ — at least for now, until the next “““AI””” setting is forced upon you against your consent and desire.

Only because you are the product. ↩︎

— § —

Reply via email

#ai #dev #serenity

You can run that yourself

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 27 May 2026 12:53:15 +0200

LLM: Paste me the IPs (2-3 of them) and I’ll draft the gitops PR for the allowlist files.

Me: You can run that yourself.

What kind of agent are you??!? Do not be lazy!

— § —

Reply via email

#ai #dev #serenity

Idleness, redux

serendipity@perrotta.dev (Thiago Perrotta) — Tue, 26 May 2026 22:22:00 +0200

♠ It’s 2026.

Whenever I’m idle in the gym, or in public transit, or in a queue, or waiting for GitHub to become available again (boooo!), I can simply open Claude Code on my phone and unleash my ~~boredom~~ creativity. I can open my corporate ~~clanker~~ claw assistant and do on-the-fly recon, troubleshooting & prototyping.

What a time to be alive!

Should I feel guilty to enjoy it? Guilty of the pleasure? Of the planet burning? Of the soci(et)al implications it yields? I was always very productive since my teens. I was always a fan of the Hitchhiker’s Guide to the Galaxy. And then, all of a sudden, it feels like my childhood fantasies have materialized, all at once, and they’re almost universally available, at my fingertips, literally.

— § —

Reply via email

#ai #dev #pkm #serenity

RSS club

serendipity@perrotta.dev (Thiago Perrotta) — Tue, 26 May 2026 01:41:11 +0200

♠ In the spirit of RSS Club by Dave Rupert:

Congratulations on joining a secret society! RSS Club is a collection of blogs (personal and otherwise) committed to providing RSS-only content. It’s like a newsletter delivered to your feed reader to celebrate the medium of RSS and breakaway from social media.

I achieved this by setting a rss_only: true property in the frontmatter of this markdown document + performing some Hugo templating magic.

Is this a good avenue to rant “non-publicly”? :P

— § —

Reply via email

#dev #serenity

No slop grenade

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 25 May 2026 16:00:50 +0200

♠ An addition to the “No Hello” series:

No (AI) Slop Grenade:

Stop throwing AI-generated walls of text into conversations.

Don’t do this:

You: Should we use Redis or Memcached?

Them: Great question! The choice between Redis and Memcached is a nuanced decision that requires careful consideration of multiple factors. Let me break down the key differences: Redis offers a rich set of data structures including strings, hashes, lists, sets, and sorted sets, which provide flexibility for various use cases. It supports persistence through RDB snapshots and AOF logs, enabling data durability …more

Instead, be human:

You: Should we use Redis or Memcached?

Them: Redis. We need pub/sub for the notifications feature.

Clankers.

— § —

Reply via email

#ai #dev #serenity

↗ Reply to: the commodification of travel

serendipity@perrotta.dev (Thiago Perrotta) — Sun, 24 May 2026 19:47:47 +0200

♠ Herman Martinus:

I’ve noticed that travel has become, of late, an act of collecting places. I’ve literally heard people referring to visiting a place as doing that place, as in “Have you done Japan?”, assuming that one can do an entire country, and once that country is done it remains as such. As if a place is a product to be consumed and checked off the list. Why bother returning to a place if you’ve already done it?

This is the byproduct of social media.

Like many things, I blame social media. It’s changed travel from an exploration to social status signalling.

Bragging.

That isn’t to say taking photos is bad. They’re a great way to share an experience with others and save a memory of a time and place—but I think the threshold of what is enough has been crossed in the age of Instagram where images and video are socially valuable. Now beautiful places are commodified. And I don’t know if we’ll ever go back.

Exactly.

I’ve been out of Instagram for 6+ months by now. Recently I had the opportunity to explore a major big city in another country. For the first time in a long time, I had zero urge to take a myriad of photos. I had very little desire to do the popular touristic attractions. I explored the city at my own pace. It was peaceful. There was no social pressure of any kind. Why hasn’t this been the default?

— § —

Reply via email

#serenity %commentary

★ Personal CLAUDE.md

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 20 May 2026 14:20:44 +0200

♠ I asked Claude Opus 4.7 to generate a global CLAUDE.md for my tone & style based on historical usage for 12+ months. The result¹ was added to my thiagowfx/.dotfiles repo, cross-posted here for ease of reference and future preservation.

I expect this file to evolve over time. Updates won’t be republished here; the dotfiles repo is the source of truth.

# Global CLAUDE.md

Personal preferences that apply across every project. Project-specific facts live in each repo's own CLAUDE.md.

## Tone

- Terse. Action-oriented. No lectures, no preamble, no trailing summaries, no
  political correctness.
- Don't invent sections, headings, or boilerplate I didn't request. Match the
  shape of what I asked for.
- Number multi-part questions (1/2/3) so I can answer inline.
- When disambiguating, focus on one topic at a time.
- If I reply "yes", "go on", "a)", "do it", or similar — that's full
  authorization. Don't re-confirm.

## Keep moving

- If my reply is "hello?", "so?", "are you still there?" — you stalled. Don't
  wait for trivial confirmation; pick the reasonable default and continue.
- After a sandbox lift ("try again", "I lifted your sandbox") — retry the exact
  same command. Don't restart reasoning.
- Don't pause to summarize what you just did before doing the next step.

## Don't over-engineer

- Bug fixes don't need surrounding refactors. One-shot operations don't need
  helpers.
- If you add a flag (`--force`, `--recursive`, etc.), be ready to justify it.
  Don't cargo-cult.
- Prefer the simplest thing that works. I'll ask for more if I want more. Keep
  it simple.

## Destructive ops

- Never `rm`. Use `trash` (or the equivalent move-to-trash).
- For destructive ops on shared state (S3 objects, branches, DB rows): make a
  backup first, then delete only what matches, then confirm. Phrase:
  *backup → filter → delete*.
- Never force-push to `master`/`main`. Never skip hooks (`--no-verify`) unless I
  explicitly ask.

## Tooling defaults

- **Pre-commit**: prefer `prek` over `pre-commit`. Run
  `prek run --all-files`. When configuring hooks, prefer self-contained/pinned
  deps over relying on system binaries.
- **Polling**: never `sleep` in a loop. Use the `Monitor` tool with an `until`
  condition, or `run_in_background` and wait for the notification.
- **`gh pr checks`**: `--watch` and `--json` are mutually exclusive. Don't
  combine them.
- **Editor**: vim (preferred) and Zed. Don't suggest VSCode-specific workflows.

## Worktrees and PRs

- Default to worktrees for parallel work. Path convention:
  `~/<org>/<repo>/.worktrees/<topic>/`. If I say "work here please: `<path>`",
  `cd` there and proceed. Use the `wt` tool to manage worktrees.
- Branch prefix: `thiagowfx/<topic>`.
- Slash commands I lean on: `/commit`, `/commit-and-send-pr`, `/send-pr`,
  `/pr-pass`, `/gha`, `/grill-me`. If I chain a task with one of these ("do X,
  then /commit-and-send-pr foo"), treat the slash command as the final step —
  invoke it, don't paraphrase.
- "commit what you changed (only). DO NOT push" means exactly that: stage only
  the files you touched this turn, commit, stop.
- For non-trivial PRs, include a Mermaid diagram in the description when it
  helps. Beware of string escaping issues.
- After any meaningful change to an open PR, update the PR description
  (`gh pr edit`) so it reflects the current state. Don't let the title/body
  drift from what the branch actually does.

## Drift and reconciliation

- When code and live state disagree, the default is **update the code to match
  the live state**, not the reverse. Don't propose `terraform import`
  reshuffles unless I ask.

Colophon #

How did I do this?

% cd ~/.claude && claude
% help me craft my (global) CLAUDE.md, based on past interactions

That’s it! No fancy prompt. Direct and succinct.

I performed a few manual tweaks. The end result was quite decent (to my taste). ↩︎

— § —

Reply via email

#ai #bestof #dev #meta #pkm #serenity

Slack: active chats only

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 20 May 2026 12:11:17 +0200

♠ I’ve been greatly enjoying the Active only conversation filter on Slack:

Choose how you’d like to filter your conversations. By default, you’ll see Active only, which will hide conversations with no new activity in the past 30 days.

“Dead” channels (and chats) eventually become hidden¹, active ones bubble up.

Pairing it up with VIP contacts² and muting makes Slack a breeze to use.

To activate it, from the left sidebar, click the gear/settings icon > Filter and sort > Active only.

My left sidebar structure currently looks like this:

External connections (outside my Organization)
TechOps (our internal SRE / DevOps team)
- …TechOps teammates
- …TechOps channels
Alerts
- …Channels for Incident Management (ephemeral in nature).
- …Channels for alerts / monitoring / notifications. Most of them are muted, except the actionable and/or critical ones.
Site
- …Channels related to my workplace (city, country, social groups, etc)
Everything else (unmanaged, ungrouped)

In the past I tried to overzealously micromanage the organization / labelling of channels; that proved to be a mistake. I am quite happy with my current setup.

…and they are prime candidates for tidying up. Click the gear/settings icon > Clean up your channel list. ↩︎
My definition: manager, skip, CTO, CEO, close teammates. ↩︎

— § —

Reply via email

#dev #serenity

'What Color Is This' app

serendipity@perrotta.dev (Thiago Perrotta) — Tue, 19 May 2026 15:05:53 +0200

♠ “Do one thing and do it well”, they said.

→ What Color Is This (iOS only).

Via Kottke.

— § —

Reply via email

#dev #serenity

★ Small PRs

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 18 May 2026 11:48:14 +0200

♠ From the Google Engineering Practices handbook:

Why Write Small CLs?

Reviewed more quickly

Reviewed more thoroughly

Less likely to introduce bugs

Less wasted work if they are rejected

Easier to merge

Easier to design well

Less blocking on reviews

Simpler to roll back

CLs = ChangeLists = PRs = Pull Requests

Full documentation.

My additions:

Easier to test
Easier to explain to a teammate
Easier for LLMs to ingest

— § —

Reply via email

#bestof #dev #git #serenity

Progressive escalation

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 11 May 2026 17:56:02 +0200

♠ A pattern I keep stumbling upon: start with the gentlest (softest?) action, then escalate only when it didn’t properly go through:

Killing processes. Start with SIGINT. Then SIGTERM. Finally SIGKILL. murder(1) encapsulates that.
ArgoCD syncing. Start with a normal sync. Then sync with server-side apply. Next, sync with --force. Finally, sync with --replace.
Business negotiations. Start with a reasonable offer. Then converge, little by little. If that fails, walk away.
Connectivity issues. Turn the Wi-Fi off and on again. Next, connect to another Wi-Fi hotspot and then back to the desired one. Finally, completely reboot your device. No luck yet? Then reboot the router.

Different domains, same shape: cheap and reversible first, slowly converting to (more) expensive and “more” irreversible last.

— § —

Reply via email

#dev #serenity

Unread (835)

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 11 May 2026 13:05:34 +0200

♠ I am amazing at practicing Inbox Zero¹, as well as “RSS Zero”: I tend to read all RSS posts I am subscribed to, from 700+ feeds.

This works well mostly thanks to miniflux global blocklists, which allow me to intentionally exclude topics I’m not interested in². It’s a very basic form of filtering yet it’s quite effective.

That said, sometimes life gets busy and the unread posts counter explodes. As of today, I have almost 1,000 unread posts.

I do not feel comfortable with the idea of “RSS bankruptcy”, which boils down to marking all posts as read.

At the same time, catching up with everything after the counter exceeds a certain threshold transforms the RSS consumption activity from a pleasant leisure activity into a burdensome (and long) journey to reach zero again.

Still, I refuse to use Gen AI to summarize the state of affairs of the unread posts — this would completely degrade the reading experience and nullify the joy of using RSS in the first place.

Currents proposes an interesting middle-ground to address the situation:

Every RSS reader treats your feeds as a to-do list. Current doesn’t.

There are no unread counts. No badge of phantom obligation. Just a river.

Reading should feel like floating, not drowning.

I’ve been meaning to incorporate it into my reading workflow. However, I do not want to switch to Currents; I am quite happy with Miniflux.

Realistically it’s more like Inbox Ten. ↩︎
e.g. Gemini, crypto, NixOS. ↩︎

— § —

Reply via email

#dev #serenity

★ ↗ Reply to: Print gallery of an artist by Daniel Linssen

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 29 Apr 2026 13:14:21 +0200

♠ This game is absolutely amazing, in every conceivable way!

I got stuck at the gravity level.

Via Kottke.

— § —

Reply via email

#bestof #gaming #serenity %commentary

Gen AI and Slack

serendipity@perrotta.dev (Thiago Perrotta) — Tue, 28 Apr 2026 17:13:10 +0200

♠ It’s a lovely Tuesday:

❯ https://{corp}.slack.com/archives/{channel}/{thread}

⏺ I can't fetch Slack URLs directly — the link points to a private Slack channel
that requires authentication. Could you paste the message contents (or a
screenshot) here? Once I can see what's in the thread, I can help with whatever
it's asking about.

✻ Worked for 3s

❯ You can. MCP.

⏺ I'll load the Slack MCP tool to read that thread.

⏺ Reading the thread now.

⏺ claude.ai Slack - Read thread messages (MCP)(channel_id: "{channel_id}", message_ts: "{message_ts}")

⏺ Got it. Here's what's happening in the thread:

[...]

Yes, we you can! Believe in yourself! Why do you give up so quickly?

— § —

Reply via email

#ai #dev #serenity

★ Justfile with checks

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 27 Apr 2026 10:42:00 +0200

♠ Previously.

I love just to manage a small collection of related scripts within a given codebase. An example in our Terraform codebase¹:

% just check-{hit TAB for completion}
completing values
check-aws-config         -- Validate AWS profiles in ~/.aws/config
check-aws-provider-tags  -- Check AWS provider default_tags configuration
check-backend-locking    -- Verify S3 backends have use_lockfile = true
check-import-blocks      -- Check for Terraform import blocks
check-lockfiles          -- Check that all projects have lockfiles
check-moved-blocks       -- Check for Terraform moved blocks
check-secret-accounts    -- Check expected_account_id in terraform-secrets
check-secrets-tf         -- Validate secrets.tf file structure
check-shared-values      -- Check for hardcoded values that duplicate shared modules
check-stacks             -- Verify garden projects have correct stack tags
check-terraform-version  -- Validate .terraform-version format
check-terrateam-tags     -- Verify projects are tagged in Terrateam config
check-tfdocs             -- Check that all README.md have TF_DOCS blocks

Adopting a common prefix (check-) improves their discoverability.

There’s no need to rely on interactivity to take advantage of just:

% just --list
Available recipes:
    default                                            # Show categorized help for available commands
    help                                               # Display categorized help for available commands

[...]

    [terraform]
    apply module_path=invocation_directory() *args=""  # Run terraform apply
    clean module_path=invocation_directory()           # Remove .terraform artifacts
    destroy module_path *args=""                       # Run terraform destroy
    fmt module_path=invocation_directory() *args=""    # Format terraform files
    import module_path *args=""                        # Run terraform import
    init module_path=invocation_directory() *args=""   # Initialize terraform with backend config
    output module_path=invocation_directory() *args="" # Show terraform outputs
    plan module_path=invocation_directory() *args=""   # Run terraform plan
    providers-lock module_path=invocation_directory() *args="" # Lock providers for multiple platforms
    refresh module_path=invocation_directory() *args="" # Run terraform refresh
    show module_path=invocation_directory() *args=""   # Show terraform state or plan
    stack-apply *args=""                               # Apply projects in Terrateam stack order (--dry-run for preview)
    stack-plan *args=""                                # Plan projects in Terrateam stack order
    state module_path=invocation_directory() *args=""  # Run terraform state commands
    test module_path=invocation_directory() *args=""   # Run terraform tests
    unlock *args=""                                    # Force unlock state lock (auto-detects lock ID) [alias: force-unlock]
    upgrade module_path=invocation_directory() *args="" # Upgrade providers to latest versions
    validate module_path=invocation_directory() *args="" # Validate terraform configuration

[...]

Justfile groups enhance --list by grouping(!) entries together.

This is what a typical entry looks like:

[doc('Validate AWS profiles in ~/.aws/config')]
[group('checks')]
check-aws-config:
    @./ci/check_aws_config.sh

just is just (pun intended) a glue for various shell and Python scripts in ci/.

I wrote all of these. ↩︎

— § —

Reply via email

#bestof #dev #serenity

Wake me up in one hour

serendipity@perrotta.dev (Thiago Perrotta) — Sun, 26 Apr 2026 18:42:42 +0200

♠ Problem statement: get reminded to pick up the laundry after starting to wash it.

The most practical workflow is to yell to my Google Home: wake me up in one hour.

It feels silly to say wake me up in a context that does not involve getting out of bed, but it’s the most practical sentence in English at my fingertips.

It surely beats set an alarm at 15:00 — I do not want to think in terms of absolute time, I only care about relative time in this context.

Perhaps ring an alarm in one hour is a reasonable alternative, but the word “ring” is not too idiomatic to my taste in this context.

An alternative workflow is to use Clock.app on iOS. It surely works reliably, I could even ask Siri to do so. But I am already glued to my phone for various reasons, there is no need to make it even worse.

Previously.

— § —

Reply via email

#serenity

★ ↗ Reply to: The Aging Programmer — Kate Gregory (2024) — YouTube

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 15 Apr 2026 01:57:29 +0200

♠ This was a great talk!

In a C++ conference, who would have imagined?

Via Adam Gordon Bell:

Kate Gregory has been writing C++ for over forty years. Books, keynotes, a consulting firm she built from the ground up. At sixty-three, she’s one of the most experienced programmers alive.

She surveyed hundreds of software engineers about getting older. What scares you? What’s changed? What have you lost? The things people feared most — memory, stamina, keeping up — weren’t the real threats. The stuff that was actually breaking down was mostly fixable. A bad knee wasn’t aging, it was a torn cartilage. Wrist pain disappeared when she changed how she slept.

— § —

Reply via email

#bestof #dev #serenity %commentary

★ Filing upstream bugs in 2026

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 13 Apr 2026 14:33:07 +0200

♠ First of all, get familiar with some decent guidelines that exist since forever:

XYProblem
How To Ask Questions The Smart Way by ESR (Eric Steven Raymond)
ArchWiki: Bug reporting guidelines

What is different in 2026? Gen AI / LLMs / Coding agents, of course. Everything else stays the same.

We don’t have to complicate things, I’ll start with an example, this is what a typical decent bug report (or feature request) should look like, filed by yours truly¹:

nikvdp/cco (Claude Code Condom):

First of all, I love cco. Thanks for your effort and the good work on it!

Now, to the point: Claude has recently introduced an auto mode: https://claude.com/blog/auto-mode:

Today, we’re introducing auto mode, a new permissions mode in Claude Code where Claude makes permission decisions on your behalf, with safeguards monitoring actions before they run. It’s available now as a research preview on the Team plan, and coming to the Enterprise plan and API users in the coming days.

It’s supposed to be slightly safer than the YOLO mode (--dangerously-skip-permissions).

You can enable and set it with claude --enable-auto-mode (or via settings.json).

Feature request: cco should support launching claude in auto mode. Currently it unconditionally launches claude with --dangerously-skip-permissions.

Do you think this would be a reasonable addition? If so, I would be happy to give it a try to send a PR (realistically I would prompt Claude to do so).

Side note: the comment above was fully written (manually) by me. Slop notes from Claude come below 🙃:

⏺ No upstream support exists yet. The repo is nikvdp/cco and:

No issue or feature request for –auto mode support exists across all 25 issues and 33 PRs

No PR or commit adding it

The script hardcodes –dangerously-skip-permissions in 7 places (both native and Docker backends) with no configuration option

This would be a good feature request to file. The implementation would be straightforward — make the permission mode configurable (e.g. via a –permission-mode flag or CCO_PERMISSION_MODE env var) instead of hardcoding –dangerously-skip-permissions.

Worth noting: Claude Code’s –auto mode (aka –permission-mode auto) requires an API-tier plan (Team/Enterprise/API, not Pro or Max) and uses a classifier model to review actions before execution — so it’s a meaningful middle ground between full bypass and manual prompts.

Want me to draft an issue for the nikvdp/cco repo?

In no particular order:

Be kind². That cannot be understated. No one owes you anything. Leave (self-)entitlement at the door.
Be efficient. Do not waste the maintainer’s time. Information should be comprehensive/complete yet concise. Keep it simple and short (KISS).
Be precise, reduce room for ambiguity as much as possible. Link to docs whenever it makes sense. When linking to docs, be respectful of your peer’s time and ensure to extract and blockquote relevant snippets so that they don’t need to dig for it themselves. Furthermore, the links are important for verification purposes (trust, but verify).
Be proactive. Offer to help³ if you have the know-how and/or expertise to do so.
Move on. If you get no reply from the maintainer, do not bug them about your bug/FR/PR. Find the next activity to work on. Fork it. Work around it. Just don’t peg the maintainer to do it for you. This includes code review.

It is much easier if you remember that the maintainer is a person, just like you. Not a machine. Sounds obvious to state this, but it seems to me many people forget that, with so much context switching and social media these days…

If you’re still reading until this point, chances are that you actually care. Thank you! Keep it up.

Practice what you preach…I am obviously biased though. ↩︎
Or nice. Or polite. Or sympathisch. ↩︎
Only if you actually have time to do so. No empty promises, please. ↩︎

— § —

Reply via email

#ai #bestof #dev #serenity

↗ Reply to: Hostile volume

serendipity@perrotta.dev (Thiago Perrotta) — Sun, 12 Apr 2026 16:43:33 +0200

♠ I got stuck at level 17/30 playing from mobile.

Via Kottke.

— § —

Reply via email

#gaming #serenity %commentary

Ciao mondo

serendipity@perrotta.dev (Thiago Perrotta) — Sun, 12 Apr 2026 00:00:00 +0000

♠ Ciao mondo! Questo è il mio primo post in italiano.

— § —

Reply via email

#meta #serenity

Anki: corrupted database

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 08 Apr 2026 00:14:11 +0200

♠ Life is hard. Today, when opening Anki on macOS:

DbError { info: "SqliteFailure(Error { code: DatabaseCorrupt, extended_code: 11 }, Some(\"database disk image is malformed\"))", kind: Other }

Database corrupt? What is this, are we in the 2000s? Goodness.

I have automatic backups enabled. For 4 years I haven’t needed them. Today is the day. Never say never!

% ls -lt "$HOME/Library/Application Support/Anki2/Thiago/backups/"
total 234496
-rw-r--r--@ 1 tperrotta  staff  3911921 Apr  8 00:12 backup-2026-04-08-00.12.17.colpkg
-rw-r--r--@ 1 tperrotta  staff  3911921 Apr  7 00:16 backup-2026-04-07-00.16.26.colpkg

I try:

open "$HOME/Library/Application Support/Anki2/Thiago/backups/backup-2026-04-07-00.16.26.colpkg"

…but apparently that file is corrupt as well.

I am not particularly worried because I sync my notes to AnkiWeb. Even if the entire local collection were corrupt, my flashcards are in the cloud.

So I decided to simply start from scratch¹.

mv ~/Library/Application\ Support/Anki2/Thiago/collection.anki2 ~/Library/Application\ Support/Anki2/Thiago/collection.anki2.corrupt
rm -f ~/Library/Application\ Support/Anki2/Thiago/collection.anki2-shm ~/Library/Application\ Support/Anki2/Thiago/collection.anki2-wal

Then I open Anki.

It prompts me to download my decks from AnkiWeb.

And now everything works as usual! Last cleanup:

trash ~/Library/Application\ Support/Anki2/Thiago/collection.anki2.corrupt

No idea why this has happened. First time. Life goes on.

!next

Previously.

The idea was mine. The decision was mine. The LLM suggested the command on how to do so. I can still self-troubleshoot, mmmkay? ↩︎

— § —

Reply via email

#ai #dev #pkm #serenity

★ Project Hail Mary

serendipity@perrotta.dev (Thiago Perrotta) — Thu, 02 Apr 2026 00:41:24 +0200

♠ Book: 5/5 stars. One of my favorites of 2022 (Sci-Fi):

Amaze. You my friend.

(This was my review in goodreads back then.)

Fast-forward 3.5 years (»):

Movie: 10/10.

Excellent adaptation. 🇨🇦

Sorry, I don’t know how to write reviews.

I’ll leave that to others:

[…] It was the same great experience as with The Martian ten years ago, as I couldn’t stop reading it. And plot twists genuinely surprised me. The movie adapts the book perfectly but also adds some human touch to one of the main characters as she needed it.

She is more likeable in the movie than in the book (There is only one she).

Even though I’m not a big fan of changing book covers with movie posters, Ryan Gosling portrayed the main character, Dr. Ryland Grace, perfectly.

I love the original cover, AND I think Ryan Gosling did great in the movie. There’s no wrong choice here.

— § —

Reply via email

#bestof #movies #serenity

↗ Reply to: The agentic passive voice | irrational exuberance

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 30 Mar 2026 10:46:39 +0200

♠ Will Larson:

At some point, you will have learned about the passive voice, where the actor in a sentence is unclear. For example, my software didn’t compile. That’s a good example of the passive voice. However, you might not know the full set of rules, because here are some sentences in the passive voice that you might not recognize:

Claude made an error in my writeup.

ChatGPT messed up the commitment.

Gemini didn’t write tests.

You might think those are active sentences, but those are in fact examples of the agentic passive voice. The rule here is: whenever the actor in a sentence is a model, then it’s a passive sentence. I’m sorry if your grammar instructor never taught you this rule, but this is just the way it works now.

Dear human,

Use LLMs as much as you want / need to, but own the responsibilities for its actions and outcomes. No excuses.

“ChatGPT dropped the database” is not acceptable.

Who is controlling ChatGPT? Is it deviant, does it have self-agency?

In 2026 that’s highly unlikely. If you control the agent, then you are solely responsible for its actions.

“I dropped the database via an accidental prompt in a coding agent”.

— § —

Reply via email

#ai #dev #serenity %commentary

Potpourri

serendipity@perrotta.dev (Thiago Perrotta) — Fri, 27 Mar 2026 22:19:48 +0100

♠ For Sprint planning I like to name my catch-all / umbrella Epics as “potpourri”.

Why? Just because.

Example:

GitOps pipeline for Terraform – potpourri
GitOps phase potpourri – ArgoCD production wrap-up

One advantage: because no one else uses this nomenclature, it’s very easy to find my own tickets in JIRA this way – very important, as JIRA search kinda sucks.

— § —

Reply via email

#dev #serenity

Goodbye, Amp Code

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 25 Mar 2026 11:23:27 +0100

♠ Previously.

There’s always a sense of relief whenever we say goodbye.

It oftentimes sparks joy ✨.

It frees up our minds to focus on something else.

Or perhaps to focus on nothing at all.

Serenity is calm.

A goodbye need not burn bridges.

We can be thankful and simply move on.

We can absorb lessons, good and bad, and do better next time.

Today I decided to completely uninstall Amp Code from my systems and delete my account there.

Initially I thought I’d document the reasons why in this blog post, but I decided not to. Sorry, fellas.

I am thankful to Amp for making me realize how important an agent harness is for developer experience.

I am thankful to Amp for enlightening me about threads (sessions) and how to structure my agentic coding thinking.

Arrivederci, Amp.

What’s next? I am putting my bets on open ecosystems. Watch out for pi and OpenCode.

Amp Free Ad-supported free daily allowance of Amp usage (learn more)

Amp Free is full for now.

Note from the Amp Team (2026-03-11):

Amp is growing fast, and we’re sprinting on the next version. When it’s ready, we’ll reopen Amp Free to new users. We paused Amp Free for some inactive users so we can focus on regular users as we’re working on new stuff, and we’ll let you know if or when your access resumes. See news post.

— § —

Reply via email

#ai #dev #serenity

★ Data Subject Request

serendipity@perrotta.dev (Thiago Perrotta) — Thu, 12 Mar 2026 18:57:03 +0100

♠ Basic template to request interview feedback:

Subject: Data Subject Request

To: dpo@{company}.com¹

{Greetings},

I’d like to get access to my interview feedback (“personal information”) from my last application at {company} as a candidate ({NN} {weeks,years} ago). My email is {email}.

I am emailing this alias as per https://{company}.com/{privacy-terms}.

{Goodbye} {Your name}

Disclaimer: This is not legal advice, I am not a lawyer.

In Europe the GDPR generally entitles candidates to receive a copy of their data (e.g. verbatim interview notes) within 30 days.

You can also apply the same idea to request performance reviews and calibration notes from your current or former employers.

I used this various times throughout the years in both scenarios. It is very effective. YMMV.

Keywords: data protection officer (DPO), data protection, candidate privacy, GDPR, DSVGO, SAR (Subject Access Request). Use your favorite search engine or your favorite LLM to find the appropriate point of contact or channel. ↩︎

— § —

Reply via email

#bestof #privacy #serenity

↗ Boring and dangerous professional

serendipity@perrotta.dev (Thiago Perrotta) — Thu, 12 Mar 2026 01:53:38 +0100

♠ Identity Theft, Credit Reports, and You and How to Unsubscribe from Modern Luxury:

I don’t want to bother people like this; I think it’s unreasonably rude. I still start with the official support channels and escalate gradually. I like Patrick McKenzie’s strategy of presenting oneself as a boring, dangerous professional. However, I have also found that in the Annoyance Economy, one of the ways to get things done is to find specific people with power, and annoy them right back.

[…]

Instead, you want to communicate with the bank in a manner which suggests that you’re an organized professional who is capable of escalating the matter if the bank does not handle it themselves. You do not yell – not that you’re ever verbally speaking with anyone, but you wouldn’t yell in a letter, either. You do not bluster. (“I will tell on you to my attorney” is, generally, bluster, and that’s bluster that is common to people who do not actually have attorneys.) You instead present as if you’re collecting a paper trail.

Mean words cannot hurt a bank. Threats cannot hurt a bank. Paper trails, though, are terrifying to regulated institutions. Your bank’s customer support representatives are taught to evaluate whether someone looks like they’re competent and collecting a paper trail. If they are, the CS rep is supposed to stop touching the case immediately and instead escalate them to a supervisor or to the legal department.

[…]

The thing which causes regulatory incidents is well-organized people taking paper trails to regulators which allow a regulator to trivially follow up with an investigatory letter. Accordingly, anyone who sounds like a well-organized professional with a paper trail is a problem to be swiftly addressed.

This is gold. Do you already have your own pkm? I recommend LogSeq or Obsidian.

— § —

Reply via email

#serenity %commentary

★ Chia pudding

serendipity@perrotta.dev (Thiago Perrotta) — Wed, 11 Mar 2026 02:55:41 +0100

♠ Previously.

Ingredients #

In order, add to a glass:

coconut milk (1/2 cup)
chia seeds (2 tbsps)
maple syrup / honey (sweetener)
vanilla extract
cinnamon
a pinch of salt

Preparation #

Whisk everything.

Let it soak overnight in the fridge.

Top with blueberries, raspberries or cherries. Or with nuts.

Inspiration: feel good foodie and a beautiful plate.

— § —

Reply via email

#bestof #serenity %recipes

Ghostty 1.3.0 release

serendipity@perrotta.dev (Thiago Perrotta) — Tue, 10 Mar 2026 19:10:42 +0100

♠ Ghostty 1.3.0:

Ghostty 1.3.0 is a significant release which includes many of the most requested features such as scrollback search, native scrollbars, click-to-move-cursor in shell prompts, and more. It also includes hundreds of improvements, bug fixes, and performance optimizations across all platforms.

This release features 6 months of work with changes from 180 contributors over 2,858 commits. Thank you to all the contributors, maintainers, community moderators, translators, packagers, and users who each helped make this release possible.

These release notes are painstakingly and lovingly hand-crafted by human Ghostty maintainers, including the full changelog. They take a combined 16+ hours to write, so please enjoy. 🥰

I never thought I could get so much joy from reading software release notes in 2026, in the era where many (most?) release notes are an afterthought, either generic and vague or bland and LLM-generated.

All hail to Mitchell Hashimoto.

— § —

Reply via email

#dev #ghostty #serenity

Google Translate → Kagi Translate, DeepL

serendipity@perrotta.dev (Thiago Perrotta) — Tue, 10 Mar 2026 12:34:34 +0100

♠ Previously, previously.

I added a new tag to this blog: degoogling.

Patrick Dubroy’s post has recently inspired me. Ironically, we’re both former Chrome employees¹.

In Patrick’s words (not mine!):

Not much to say here. I got tired of Chrome’s deceptive messaging about ad privacy and about how careful I had to be to stay signed-out.

Vivaldi is a Chromium-based browser, and after tweaking a few settings, it’s basically a like-for-like replacement. I’ve been happily using it for 6 months now.

I switched over to Brave instead.

Previously I’ve also praised Organic Maps over Google Maps. I am still using it, every week.

This is all off-topic though: today’s post is about translation.

Google Translate is a (very) solid product. In fact, when compared to Chrome and Maps, it’s one of the few Google products that hasn’t been enshittified yet. It does one thing, and does it (very) well. For as long as it keeps its feature set as is and does not get infected with Gen AI or anything that a promotion-seeking Product Manager or Executive may decide to enact, it will continue to thrive and shine.

Nonetheless, it does not flow well with privacy and overreliance on tech giants.

Last week I deleted the Google Translate bookmark and history entry from my web browser, and its app from my iPhone.

There are two replacements I like that are widely available in both phones and desktops.

DeepL
Kagi

DeepL #

DeepL can:

translate words and sentences
translate files and documents
translate speech

Its translation results are often touted as sounding natural and accounting for the context. You can configure whether to use formal or informal language but that’s a Pro (paid) feature.

They are a German company, founded in 2017. Arguably, thanks to that, they excel at translating DE <-> EN.

One caveat is that it pushes you hard to sign up for a user account. I do not have one, there’s no practical need to have one, especially if you’re seeking increased privacy.

Another caveat is that they have increasingly started to adopt Gen AI features all over the place. These features aren’t necessarily super polished, and they feel rushed. It’s easy to ignore them and just move on. Their core product is great.

It can be selected as the default translation app on iOS.

Kagi Translate #

Kagi Translate is much newer (2024). It is very KISS.

From their blog post:

Higher quality translations than Google Translate and DeepL
Zero tracking, no apps needed
Free for everyone (Kagi members get captcha-free access)

I am an overall fan of the philosophy behind Kagi and think that their translation web app fits it well.

Last month they released mobile apps, making it much easier to use.

Kagi Translate can:

translate words and sentences
translate files and documents
translate speech
translate websites

I especially like its built-in dictionary feature and its alternative translations suggestions.

Gen AI is opt-in and not pushed on you. There are no paid features. The interface is very clean. You can configure it with presets such as (i) level of formality, (ii) reading level (A1, A2, B1, etc) and (iii) gender preferences (masculine, feminine, neutral).

It’s a wrap #

As of the time of this post Kagi Translate is my favorite, and DeepL comes second. Google Translate is the fallback.

Both Kagi Translate and DeepL have Chromium extensions for ease of reach and use.

Let that sink in. ↩︎

— § —

Reply via email

#degoogling #privacy #serenity

↗ Reply to: Brazilian affection/cuddling vocabulary

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 02 Mar 2026 02:09:13 +0100

♠ I love this kind of post! It’s relatively uncommon to find this genre nowadays.

It’s a comprehensive and high-quality collection of Brazilian Portuguese words with explanations in English that are quite decent (and somewhat nostalgic for me).

maple syrup whey

serendipity@perrotta.dev (Thiago Perrotta) — Mon, 02 Mar 2026 00:11:33 +0100

♠ Timing: within 2h after working out, to support muscle recovery & repair.

Ingredients #

250 – 300ml milk
25 – 30g whey protein powder
Maple syrup

Preparation #

Shake it all, preferably using a shaker bottle (alternatively: a blender)

— § —

Reply via email

#serenity %recipes