# Blogger Automation Research

This is a source survey for bloggers whose sites behave like personal infrastructure: custom build systems, script libraries, link preservation, publishing checks, AI helpers, and command-line workflows.

## Gwern.net

Sources:

- Local source: `/home/krisyotam/src/parsing/gwern.net/build/bash.sh`
- Local source: `/home/krisyotam/src/parsing/gwern.net/build/sync.sh`
- Local source: `/home/krisyotam/src/parsing/gwern.net/build/linkArchive.sh`
- https://gwern.net/Design
- https://gwern.net/archiving
- https://gwern.net/static/build/

Patterns worth copying:

- A sourced personal shell library for site work, not just a Makefile. Gwern's `bash.sh` has path conversion, colored check wrappers, Cloudflare cache expiry, PDF surgery, image composition, search helpers, rename helpers, tag helpers, and upload aliases.
- A full sync script that behaves like a release engineer: dependency checks, disk-space checks, cleanup, process collision checks, priority lowering, infrastructure pulls, global text rewrites, compilation, metadata checks, backlink generation, sitemap work, link checking, and upload.
- A rich publishing helper. Gwern's `upload.sh` lowercases filenames, normalizes extensions, converts media, OCR/compresses PDFs, strips metadata, handles name collisions, rsyncs, previews, adds small files to git, and keeps large files out of git.
- Preemptive link archiving. `LinkArchive.hs` and `linkArchive.sh` map URL to stable local path by domain plus SHA1, use SingleFile for HTML, treat PDFs specially, OCR/compress PDFs, preview snapshots, and split huge single-file archives.
- Link extraction as an archiving queue: `find . -name "*.md" -type f -print0 | xargs -0 linkExtractor | filter-urls >> ~/.urls.txt`.
- Style as code: global rewrites for URLs, spelling, citation conventions, typography, link cruft, and metadata normalization.
- Golden regression pages. Download known stress-test pages, strip cache-busting query strings, and diff rendered HTML against snapshots.
- Time-gated checks. A helper like `everyNDays 7 && expensive-audit` keeps the default build fast.

Reusable commands:

```sh
blog-links content/**/*.md | sort -u > archive-queue.txt
while read -r url; do blog-archive-url "$url"; done < archive-queue.txt
blog-frontmatter
blog-linkrot content/**/*.md
blog-backlinks > backlinks.md
```

## Simon Willison

Sources:

- https://simonwillison.net/about/
- https://simonwillison.net/series/blogging/
- https://datasette.io/for/websites
- https://github.com/simonw/simonw
- https://github.com/simonw/shot-scraper
- https://github.com/simonw/files-to-prompt
- https://llm.datasette.io/en/stable/templates.html
- https://llm.datasette.io/en/latest/usage.html
- https://tools.simonwillison.net/

Patterns worth copying:

- Treat the blog as an application with a database. Simon's site is Django backed by PostgreSQL, exported to JSON backups, then published through Datasette surfaces.
- Publish structured data as a first-class artifact. Datasette can turn SQLite files into searchable, filterable public websites and APIs.
- Keep many feeds: all content, long posts, links, quotations, per-tag feeds, and series feeds.
- Use AI from the command line through reusable templates and fragments, with prompts saved as named tools rather than pasted one-offs.
- Publish small web tools with source links, commit history, and transcripts where useful.
- Generated sections with stable markers. Simon's GitHub profile README bot replaces marked chunks from feeds, GraphQL, and Datasette JSON, then commits only when content changes.
- Screenshot automation. `shot-scraper multi shots.yml` plus GitHub Actions can archive visual states after deploys.
- Corpus-to-prompt helpers. `files-to-prompt` bundles selected files into clean model context.

Reusable commands:

```sh
llm --system 'Review this draft for weak claims and missing sources' --save blog-review
cat draft.md | llm -t blog-review
files-to-prompt content -e md -e mdx --markdown | llm -s 'Find stale claims and missing sources'
shot-scraper https://example.com/ -o screenshots/home.png
sqlite-utils insert blog.db posts posts.json --pk id
datasette publish cloudrun blog.db -m metadata.json
```

## Andy Matuschak

Sources:

- https://notes.andymatuschak.org/About_these_notes
- https://notes.andymatuschak.org/Evergreen_notes
- https://github.com/andymatuschak/note-link-janitor
- https://github.com/andymatuschak/orbit

Patterns worth copying:

- Notes as public working memory, not just finished articles.
- Atomic notes, dense links, and repeated revision beat chronological publication alone.
- Treat note status as a first-class field. The 100x scripts should index drafts, flags, and backlinks rather than assuming everything is a finished post.
- Idempotent generated backlinks for `[[wiki links]]`: `note-link-janitor path/to/notes`.
- Notes can compile into interactive artifacts. Orbit's note-sync pattern treats Markdown prompts as source material for spaced review.

Reusable commands:

```sh
blog-draft-index
blog-backlinks > backlinks.md
blog_find 'evergreen|atomic|status'
```

## Maggie Appleton

Sources:

- https://maggieappleton.com/garden-history
- https://github.com/MaggieAppleton/digital-gardeners
- https://github.com/MaggieAppleton/maggieappleton.com-V2
- https://github.com/MaggieAppleton/maggieappleton.com-V3

Patterns worth copying:

- Digital garden maturity states. Drafts can be marked as seed, budding, evergreen, archived, or private instead of a binary published/draft flag.
- Visual and conceptual metadata matter. Garden pages use state, topic, and relationship as interface primitives.
- Keep a public list of other gardens as research fodder.
- Content-derived garden indexes. Maggie's V3 scripts generate links, topics, webmentions, and small "smidgeon" drafts before Astro preview/build.

Reusable commands:

```sh
rg -n 'status:|growth:|stage:' content notes garden
blog-frontmatter
blog-draft-index
```

## Tom MacWright

Sources:

- https://macwright.com/writing
- https://tmcw-ama.val.run/
- https://macwright.com/2023/12/14/blog-about-blog
- https://macwright.com/2020/05/02/linkrot
- https://macwright.com/2025/04/03/personal-tools
- https://github.com/tmcw/notfoundbot

Patterns worth copying:

- Simple, fast, personal site with a clear stance on writing and tooling.
- Public archive pages and "recently" posts as a low-friction cadence.
- Explicit disclosure about AI usage. If AI is part of the writing pipeline, make that policy visible and use scripts for review, not silent authorship.
- Linkrot repair as PRs, not silent rewrites. `notfoundbot` can upgrade HTTP to HTTPS, find Wayback captures, and propose small diffs.
- Markdown AST plus `MagicString` minimal URL rewrites is the right shape for future citation/link tools.
- Tiny personal desk tools around existing CLIs, for example `gh search prs ... | jq ...`, translate well to draft queues and publish checklists.

Reusable commands:

```sh
blog-new "Recently"
blog-stats
blog-ai-review draft.md
```

## Robin Sloan

Sources:

- https://www.robinsloan.com/lab/claude-revision/
- https://gist.github.com/robinsloan/6d5f4247b5f2239872bea8391fd63016

Patterns worth copying:

- AI revision loops should be auditable. Sloan's script preserves the first plan and every revision file, making each model pass inspectable.
- Separate "plan the revision" from "apply the revision" so model edits can be evaluated rather than blindly accepted.

Reusable commands:

```sh
blog-ai-review draft.md > review.txt
cp draft.md draft.v1.md
```

## Derek Sivers

Sources:

- https://code.sivers.org/
- https://data.sivers.org/uses
- https://wiki.sive.rs/plaintext
- https://sive.rs/7
- https://code.sivers.org/ws
- https://github.com/sivers/sivers

Patterns worth copying:

- Plain text as durable source of truth.
- Site as personal database: articles, book notes, translations, now pages, project pages, and direct sales.
- Terminal-first stack: Vim, PostgreSQL, Ruby/Go, Sinatra, OpenBSD, mutt, Beancount, sox, ffmpeg, mpv, mupdf, ssh.
- Public code and public ideas as an operating principle.
- Unusual architecture: Go servers route requests to PostgreSQL functions that generate HTML and headers. `scripts/tap` wraps pgTAP tests in rollback transactions.
- Clear anti-AI disclosure: "Everything here is written by me, personally, not an AI."

Reusable commands:

```sh
blog-new "Now"
blog_words
blog_find 'plain text'
```

## Brendan Gregg

Sources:

- https://github.com/brendangregg
- https://github.com/brendangregg/FlameGraph
- https://github.com/brendangregg/perf-tools
- https://cacm.acm.org/practice/the-flame-graph/

Patterns worth copying:

- Posts that ship with reusable diagnostic scripts become long-lived references.
- Visualization generators are publishing tools: data in, SVG out, explanation around it.
- Keep a script archive next to explanations so readers can reproduce the picture.
- Keep intermediate folded/text artifacts because they are grep-able and reusable.
- Test visualization scripts with fixture inputs, collapsed expected outputs, and SVG generation.

Reusable commands:

```sh
python -m cProfile -o output.prof script.py
flameprof output.prof --format log > output.folded
flamegraph.pl output.folded > profile.svg
```

## Org Mode Static-Site Builders

Sources:

- https://ogbe.net/blog/emacs_org_static_site
- https://github.com/jethrokuan/braindump
- https://github.com/freetonik/rakhim.org
- https://github.com/freetonik/braindump.rakhim.org
- https://github.com/sachac/.emacs.d

Patterns worth copying:

- A self-contained private source repo with `bib`, `bin`, `blog`, `css`, `cv`, `html`, `img`, `lisp`, and `pages`.
- `make` plus Emacs batch mode as a static-site compiler.
- Custom export backend, sitemap generator, CSS/JS minification, MathJax hosting, publication lists from BibTeX, and preprocessing/postprocessing passes.
- Jethro Kuan generates `build.ninja` from `build.py`, then runs many Emacs batch Org exports in parallel before Hugo builds.
- Rakhim's braindump pipeline is explicit: `clean copy heading backlinks wikilinks frontmatter build publish`.
- Sacha Chua uses literate Emacs publishing: Make runs Emacs batch to tangle `Sacha.org` into `Sacha.el`, export HTML, and create backlinks from generated files to source sections.

Reusable commands:

```sh
make
emacs --batch -l lisp/build-site.el -f org-publish-all
blog-rss-check public/feed.xml
```

## Patrick McKenzie

Sources:

- https://www.kalzumeus.com/
- https://gist.github.com/patio11

Patterns worth copying:

- Publish narrow, production-adjacent gists when the main site code is private.
- Export Markdown versions of posts for readability, LLM context, or archival use, with allowlisted frontmatter and per-post opt-outs.
- Publish hashes for important public claims or documents.

## Dan Luu

Sources:

- https://danluu.com/octopress-speedup/
- https://danluu.com/about/
- https://gist.github.com/danluu

Patterns worth copying:

- Performance posts should be reproducible: WebPageTest runs, screenshots, each intervention, each latency delta.
- Minimal static-site maintenance is a scriptable discipline: remove unused JS, move blocking JS to the footer, combine/minify assets, remove webfonts, consider inline CSS for single-entry visits.
- Maintain lightweight bibliographies as public gists.

## Julia Evans

Sources:

- https://jvns.ca/blog/2016/10/09/switching-to-hugo/
- https://jvns.ca/blog/2017/12/22/how-i-set-up-an-automatic-weekly-blog-digest/
- https://jvns.ca/blog/2021/11/15/esbuild-vue/
- https://github.com/jvns

Patterns worth copying:

- Keep a tiny post generator task, such as a Rake `new_post`, so filenames and frontmatter are never hand-typed.
- Prefer durable static binaries for authoring tools when language environments become fragile.
- Use single-command small frontend builds: `esbuild script.js --bundle --minify --outfile=bundle.js`.
- Debug toolchain weirdness with system tools such as `strace -e openat -f`.

## Xe Iaso

Sources:

- https://github.com/Xe/site
- https://github.com/Xe/site/blob/main/scripts/validate-blog-dates.js
- https://github.com/Xe/site/blob/main/scripts/check-mdx-tags.go
- https://github.com/Xe/site/blob/main/Earthfile

Patterns worth copying:

- Document agent/developer entrypoints in `AGENTS.md`.
- Future-dated content guard: block future dates unless the PR body contains an explicit "DO NOT MERGE until YYYY-MM-DD UTC" line.
- Validate MDX tags with a small dedicated checker.
- Keep diagrams as source plus output, for example `.dot` beside `.svg`.

Reusable commands:

```sh
blog-future-dates
BLOG_ALLOW_FUTURE_DATES=1 blog-future-dates
```

## Eli Bendersky

Sources:

- https://eli.thegreenplace.net/pages/about
- https://github.com/eliben/code-for-blog

Patterns worth copying:

- Keep a separate `code-for-blog` repo organized by year and article.
- Article directories should include runnable examples, Makefiles, tests, benchmarks, generated plots, and README notes.
- For diagram-heavy or benchmark-heavy posts, publish generation scripts, not only final images.

## Buster Benson

Sources:

- https://github.com/busterbenson/public

Patterns worth copying:

- Treat a personal canon as a repo: beliefs, ideas, changelog, yearly review commits, and public feedback.
- The "public accountable self" idea maps cleanly to versioned `beliefs.md`, `ideas.md`, and annual review indexes.

## Nikita Voloboev

Sources:

- https://github.com/nikivdev/go

Patterns worth copying:

- `flow.toml` as a task-runner control plane with commands for setup, dev, spec printing, deploy, and AI-backed commits.
- Build CLIs to temporary directories, run them, clean up, and install short symlink aliases.
- Generate prompt text from Markdown specs with a command like `spec-print note.md --stdout --no-clipboard`.

## Phil Eaton

Sources:

- https://github.com/eatonphil/eatonphil.com

Patterns worth copying:

- A one-file Go static-site compiler is enough for many personal sites.
- Emit per-section output, sitemap, and robots files without adopting a large framework.

## Linus Lee

Sources:

- https://github.com/thesephist/stream
- https://github.com/thesephist/sistine

Patterns worth copying:

- JSONL is a good microblog database.
- `make watch` with `entr` and a tiny formatter target gives a fast authoring loop.
- A small custom static-site generator can be maintained like any other personal tool.

## Cross-Cutting Ideas

- `new`: create a dated draft with title, status, tags, canonical slug.
- `index`: list drafts by age, word count, status, and TODO flags.
- `links`: extract all external links with source file references.
- `archive`: save every external link locally before publication.
- `title`: fetch a page title and normalize it for citation notes.
- `frontmatter`: fail fast on missing title/date/status/tags.
- `rss`: validate Atom/RSS as part of the build.
- `linkrot`: run a polite status checker on changed posts.
- `backlinks`: generate a reverse-link report for notes.
- `stats`: count posts, words, links, domains, and TODO flags.
- `ai-review`: use a CLI LLM as a skeptical editor, never as silent author.
- `build-check`: run whatever the repo provides: Makefile, npm, Hugo, Jekyll, frontmatter, RSS, stats.
- `future-dates`: block scheduled posts unless explicitly allowed.
- `screenshot`: archive visual states after deploys.
- `spec-print`: turn Markdown notes/specs into reusable AI prompts.
- `wikilinks`: convert `[[Page]]` and `![[file.png]]` before build.
- `tangle-build`: tangle literate source and publish generated HTML/PDF.
- `db-test`: run database-backed site tests inside rollback transactions.