States/Context/Embeddings fixes

.env.example

@@ -1,9 +1,15 @@
-# Copy this file to .env and fill in the values.
+# Copy this file to .env and fill in the values, then run:
-# .env is gitignored — never commit it.
+#     chmod 600 .env
+# so other local users cannot read your VPS path / token. .env is
+# gitignored — never commit it. The auto-snapshot in `make build`
+# uses an explicit pathspec under content/ to keep stray .env files
+# out of the snapshot, but **/.env is also in .gitignore as a backstop.
 #
-# `make deploy` rsyncs the built _site/ to the VPS, then pushes the
+# `make deploy` pushes to GitHub first, then rsyncs the built _site/
-# repository to GitHub. The Makefile aborts with a clear error if any
+# to the VPS. The Makefile aborts with a clear error if any of
-# of VPS_USER / VPS_HOST / VPS_PATH is unset.
+# VPS_USER / VPS_HOST / VPS_PATH is unset, if VPS_PATH points at an
+# obviously dangerous parent directory, or if _site/index.html does
+# not exist (a sign of a broken build).
 
 # --- VPS deployment target -------------------------------------------------
 # SSH user on the deployment VPS.
@@ -15,8 +21,10 @@ VPS_PATH=
 
 # --- GitHub mirror push ----------------------------------------------------
 # A GitHub fine-grained personal access token with Contents: read+write
-# on the levineuwirth.org repository.
+# on the levineuwirth.org repository. Currently optional — `make deploy`
-# Generate at: https://github.com/settings/tokens
+# uses your local git credential helper for `git push`, so this is only
+# needed if you wire token-based push into a credential helper yourself.
+# Generate at: https://github.com/settings/personal-access-tokens/new
 GITHUB_TOKEN=
 
 # The GitHub repository in owner/repo format.

.gitignore

@@ -3,6 +3,10 @@ _site/
 _cache/
 .DS_Store
 .env
+# Defense-in-depth: catch any stray .env / .env.* anywhere in the tree
+# (the auto-snapshot in the Makefile stages content/ on every build).
+**/.env
+**/.env.*
 
 # Editor backup/swap files
 *~

Makefile

@@ -1,9 +1,12 @@
 .PHONY: build deploy sign download-model download-pdfjs compress-assets convert-images pdf-thumbs pdfs watch clean dev
 
-# Source .env for GITHUB_TOKEN and GITHUB_REPO if it exists.
+# Source .env for deploy / GitHub config if it exists.
 # .env format: KEY=value (one per line, no `export` prefix, no quotes needed).
+# Only the variables explicitly listed below are exported to recipe
+# subprocesses — bare `export` would leak every .env key (including any
+# future GITHUB_TOKEN) into every child process.
 -include .env
-export
+export VPS_USER VPS_HOST VPS_PATH GITHUB_REPO
 
 build:
 	# Auto-snapshot any uncommitted content/ changes BEFORE the build
@@ -12,8 +15,20 @@ build:
 	# the history — that's intentional. The next successful build
 	# either reuses it (no new content/ changes) or appends another
 	# snapshot on top, so failures don't disappear from the log.
-	@git add content/
+	#
-	@git diff --cached --quiet || git commit -m "auto: $$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+	# Pathspec is explicit (not `git add content/`) so a stray .env,
+	# credential file, or other non-content artifact dropped under
+	# content/ is NOT auto-staged. The :(glob) magic prefix makes `**`
+	# match across path components (git default fnmatch does not).
+	# Add new extensions here if a new asset type is introduced.
+	@git add ':(glob)content/**/*.md' ':(glob)content/**/*.html' ':(glob)content/**/*.bib' \
+	         ':(glob)content/**/*.png' ':(glob)content/**/*.jpg' ':(glob)content/**/*.jpeg' \
+	         ':(glob)content/**/*.svg' ':(glob)content/**/*.gif' ':(glob)content/**/*.pdf' \
+	         ':(glob)content/**/*.mp3' ':(glob)content/**/*.ogg' ':(glob)content/**/*.flac' \
+	         ':(glob)content/**/*.yaml' ':(glob)content/**/*.yml' ':(glob)content/**/*.json' \
+	         ':(glob)content/**/*.css' ':(glob)content/**/*.tex'
+	@git diff --cached --quiet || git commit -m "auto: $$(date -u +%Y-%m-%dT%H:%M:%SZ) [skip ci]"
+	@mkdir -p data
 	@date +%s > data/build-start.txt
 	@./tools/convert-images.sh
 	@$(MAKE) -s pdf-thumbs
@@ -29,7 +44,8 @@ build:
 	> IGNORE.txt
 	@BUILD_END=$$(date +%s); \
 	 BUILD_START=$$(cat data/build-start.txt); \
-	 echo $$((BUILD_END - BUILD_START)) > data/last-build-seconds.txt
+	 echo $$((BUILD_END - BUILD_START)) > data/last-build-seconds.txt.tmp && \
+	 mv data/last-build-seconds.txt.tmp data/last-build-seconds.txt
 
 sign:
 	@./tools/sign-site.sh
@@ -99,9 +115,19 @@ deploy: clean build sign
 	@test -n "$(VPS_USER)" || (echo "deploy: VPS_USER not set in .env" >&2; exit 1)
 	@test -n "$(VPS_HOST)" || (echo "deploy: VPS_HOST not set in .env" >&2; exit 1)
 	@test -n "$(VPS_PATH)" || (echo "deploy: VPS_PATH not set in .env" >&2; exit 1)
-	@command -v notify-send >/dev/null 2>&1 && notify-send "make deploy" "Ready to rsync — waiting for SSH auth" || true
+	# Refuse to deploy a manifestly broken build. _site/index.html must
-	rsync -avz --delete _site/ $(VPS_USER)@$(VPS_HOST):$(VPS_PATH)/
+	# exist and be non-empty before we run rsync --delete on the VPS.
+	@test -s _site/index.html || { echo "deploy: _site/index.html is missing or empty — refusing to rsync" >&2; exit 1; }
+	# Defense-in-depth: refuse rsync --delete to obviously dangerous
+	# parents in case VPS_PATH was typo'd (e.g. trailing-slash mistake).
+	@case "$(VPS_PATH)" in /|/srv|/srv/http|/var|/var/www|/home|/root|"") echo "deploy: VPS_PATH=$(VPS_PATH) looks unsafe — refusing" >&2; exit 1 ;; esac
+	@command -v notify-send >/dev/null 2>&1 && notify-send "make deploy" "Ready to push & rsync — waiting for auth" || true
+	# Push first: a successful push is cheap to roll back, while a
+	# half-completed rsync is harder to recover from. If the push
+	# fails (auth, branch protection, network), abort before touching
+	# the VPS so the public source repo and the live site stay in sync.
 	git push -u origin main
+	rsync -avz --delete _site/ $(VPS_USER)@$(VPS_HOST):$(VPS_PATH)/
 
 watch: export SITE_ENV = dev
 watch:
@@ -117,4 +143,4 @@ dev: export SITE_ENV = dev
 dev:
 	cabal run site -- clean
 	cabal run site -- build
-	python3 -m http.server 8000 --directory _site
+	python3 -m http.server 8000 --bind 127.0.0.1 --directory _site

build/Contexts.hs

@@ -10,6 +10,7 @@ module Contexts
     , compositionCtx
     , contentKindField
     , abstractField
+    , descriptionField
     , tagLinksField
     , tagLinksFieldExcludingScope
     , tagLinksFieldExcludingTopSegment
@@ -34,7 +35,7 @@ import Data.Time.Format        (formatTime, defaultTimeLocale, parseTimeM)
 import System.FilePath         (takeDirectory, takeFileName)
 import Text.Read               (readMaybe)
 import qualified Data.Text     as T
-import Text.Pandoc             (runPure, readMarkdown, writeHtml5String, Pandoc(..), Block(..), Inline(..))
+import Text.Pandoc             (runPure, readMarkdown, writeHtml5String, writePlain, Pandoc(..), Block(..), Inline(..))
 import Text.Pandoc.Options     (WriterOptions(..), HTMLMathMethod(..))
 import Hakyll       hiding (trim)
 import Backlinks    (backlinksField)
@@ -348,6 +349,44 @@ abstractField = field "abstract" $ \item -> do
     isPara (Para _) = True
     isPara _        = False
 
+-- ---------------------------------------------------------------------------
+-- Description field
+-- ---------------------------------------------------------------------------
+
+-- | Renders the @abstract@ frontmatter key as plain text suitable for use in
+--   @<meta name="description">@, @og:description@, and @twitter:description@.
+--   Strips Pandoc markup, collapses internal whitespace, truncates to ~200
+--   chars, and HTML-escapes attribute-special characters. Returns @noResult@
+--   when no @abstract@ is present (so @$if(description)$@ short-circuits).
+descriptionField :: Context String
+descriptionField = field "description" $ \item -> do
+    meta <- getMetadata (itemIdentifier item)
+    case lookupString "abstract" meta of
+        Nothing  -> fail "no abstract"
+        Just src -> do
+            let pandocResult = runPure $ do
+                    doc <- readMarkdown defaultHakyllReaderOptions (T.pack src)
+                    writePlain defaultHakyllWriterOptions doc
+            case pandocResult of
+                Left err  -> fail $ "Pandoc error rendering description: " ++ show err
+                Right txt ->
+                    let collapsed = T.unwords (T.words txt)
+                        capped    = if T.length collapsed > 200
+                                       then T.take 197 collapsed <> T.pack "\x2026"
+                                       else collapsed
+                    in  return (attrEscape (T.unpack capped))
+
+-- | HTML-escape characters that would break out of an attribute value.
+attrEscape :: String -> String
+attrEscape = concatMap esc
+  where
+    esc '&'  = "&amp;"
+    esc '<'  = "&lt;"
+    esc '>'  = "&gt;"
+    esc '"'  = "&quot;"
+    esc '\'' = "&#39;"
+    esc c    = [c]
+
 -- ---------------------------------------------------------------------------
 -- Summary field
 -- ---------------------------------------------------------------------------
@@ -377,6 +416,7 @@ siteCtx =
     <> buildTimeField
     <> pageScriptsField
     <> abstractField
+    <> descriptionField
     <> summaryField
     <> dingbatField
     <> defaultContext

build/Stats.hs

@@ -619,7 +619,10 @@ renderDistribution wcs =
              ]
     counts = foldr (\w acc -> Map.insertWith (+) (bucketOf w) (1 :: Int) acc)
                    (Map.fromList [(i, 0 :: Int) | i <- [0 .. 4]]) wcs
-    buckets  = [(labels !! i, fromMaybe 0 (Map.lookup i counts)) | i <- [0 .. 4]]
+    -- Pair labels with bucket indices via @zip@ rather than @(!!)@ to keep
+    -- the function total even if the bucket count and @labels@ list ever
+    -- drift out of sync (matching the discipline used in 'median').
+    buckets = [(lbl, fromMaybe 0 (Map.lookup i counts)) | (i, lbl) <- zip [0 :: Int ..] labels]
     maxCount = max 1 (maximum (map snd buckets))
     bar (lbl, n) =
         let pct = n * 100 `div` maxCount

content/essays/empirical-musings-spaced-repetition.md

@@ -17,6 +17,6 @@ evidence: 2
 scope: broad
 novelty: idiosyncratic
 practicality: high
-confidence history:
+confidence-history:
    - 65
 ---

content/essays/notes-from-underground.md

@@ -8,7 +8,7 @@ tags:
   - nonfiction/philosophy
 authors:
   - "Levi Neuwirth | /me.html"
-revised:
+history:
   - date: "2026-04-17"
     note: "expanded section on Shestov's divergence from Nietzsche"
   - date: "2025-12-03"

content/library.md

@@ -1,3 +1,8 @@
+---
+title: Library
+library: true
+---
+
 ::: {lang="es"}
 > *El universo (que otros llaman la Biblioteca) se compone de un número indefinido, y tal vez infinito, de galerías hexagonales, con vastos pozos de ventilación en el medio, cercados por barandas bajísimas.*
 >

static/js/popups.js

@@ -294,20 +294,32 @@
     }
 
     /* 1. Citations — synchronous DOM lookup; supports multi-citation groups
-          via data-cite-keys (space-separated list of ref-* IDs). */
+          via data-cite-keys (space-separated list of ref-* IDs).
+          Returns a DocumentFragment of cloned bibliography entries instead
+          of stringifying innerHTML, so a malicious or malformed cite target
+          cannot smuggle markup through the popup's innerHTML setter. */
     function citationContent(target) {
         return new Promise(function (resolve) {
             var keysAttr = target.getAttribute('data-cite-keys');
             var ids = keysAttr
                 ? keysAttr.trim().split(/\s+/)
                 : [(target.getAttribute('href') || '').slice(1)];
-            var parts = ids.map(function (id) {
+            var entries = ids
-                var entry = document.getElementById(id);
+                .map(function (id) { return document.getElementById(id); })
-                return entry ? '<div class="popup-citation-entry">' + entry.innerHTML + '</div>' : null;
+                .filter(Boolean);
-            }).filter(Boolean);
+            if (!entries.length) { resolve(null); return; }
-            resolve(parts.length
+
-                ? '<div class="popup-citation">' + parts.join('') + '</div>'
+            var wrapper = document.createElement('div');
-                : null);
+            wrapper.className = 'popup-citation';
+            entries.forEach(function (entry) {
+                var item = document.createElement('div');
+                item.className = 'popup-citation-entry';
+                Array.prototype.forEach.call(entry.childNodes, function (n) {
+                    item.appendChild(n.cloneNode(true));
+                });
+                wrapper.appendChild(item);
+            });
+            resolve(wrapper);
         });
     }
 

templates/partials/footer.html

@@ -4,7 +4,7 @@
         <a href="/">levineuwirth.org</a>
     </div>
     <div class="footer-center">
-        <span class="footer-license">CC BY-SA-NC 4.0 · <a href="https://git.levineuwirth.org/neuwirth/levineuwirth.org">MIT</a> · <a href="/memento-mori.html" class="footer-mm">MM</a></span>
+        <span class="footer-license"><a href="https://creativecommons.org/licenses/by-nc-sa/4.0/" rel="license">CC&nbsp;BY-NC-SA&nbsp;4.0</a> · <a href="https://git.levineuwirth.org/neuwirth/levineuwirth.org">MIT</a> · <a href="/memento-mori.html" class="footer-mm">MM</a></span>
     </div>
     <div class="footer-right">
         <a href="/build/" class="footer-build-link" aria-label="Build telemetry">build</a> $build-time$

templates/partials/head.html

@@ -1,6 +1,21 @@
-<meta charset="UTF-8">
+<meta charset="utf-8">
-<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<meta name="viewport" content="width=device-width, initial-scale=1">
-$if(home)$<title>Levi Neuwirth</title>$else$<title>$title$ — Levi Neuwirth</title>$endif$
+$if(home)$<title>Levi Neuwirth</title>$else$$if(title)$<title>$title$ — Levi Neuwirth</title>$else$<title>Levi Neuwirth</title>$endif$$endif$
+$if(description)$<meta name="description" content="$description$">$endif$
+<link rel="canonical" href="$site-url$$url$">
+<link rel="alternate" type="application/atom+xml" title="Levi Neuwirth" href="/feed.xml">
+<link rel="alternate" type="application/atom+xml" title="Levi Neuwirth — music" href="/music/feed.xml">
+
+<!-- OpenGraph / Twitter (link-preview unfurling) -->
+<meta property="og:site_name" content="Levi Neuwirth">
+$if(home)$<meta property="og:title" content="Levi Neuwirth">$else$$if(title)$<meta property="og:title" content="$title$">$endif$$endif$
+$if(description)$<meta property="og:description" content="$description$">$endif$
+<meta property="og:url" content="$site-url$$url$">
+$if(date)$<meta property="og:type" content="article">$else$<meta property="og:type" content="website">$endif$
+<meta property="og:image" content="$site-url$/web-app-manifest-512x512.png">
+<meta name="twitter:card" content="summary">
+$if(description)$<meta name="twitter:description" content="$description$">$endif$
+
 <link rel="icon" type="image/png" href="/favicon-96x96.png" sizes="96x96">
 <link rel="icon" type="image/svg+xml" href="/favicon.svg">
 <link rel="shortcut icon" href="/favicon.ico">

tools/embed.py

@@ -16,6 +16,7 @@ Staleness check: skips if all output files are newer than every HTML in _site/.
 """
 
 import json
+import os
 import re
 import sys
 from pathlib import Path
@@ -45,6 +46,19 @@ MAX_PARA_CHARS = 1000   # semantic: truncate before embedding
 
 EXCLUDE_URLS = {"/search/", "/build/", "/404.html", "/feed.xml", "/music/feed.xml"}
 
+
+def atomic_write_bytes(path: Path, data: bytes) -> None:
+    """Write to path.tmp then os.replace, so an interrupt mid-write
+    cannot leave a truncated file that the next build/serve loads."""
+    path.parent.mkdir(parents=True, exist_ok=True)
+    tmp = path.with_suffix(path.suffix + ".tmp")
+    tmp.write_bytes(data)
+    os.replace(tmp, path)
+
+
+def atomic_write_text(path: Path, text: str) -> None:
+    atomic_write_bytes(path, text.encode("utf-8"))
+
 STRIP_SELECTORS = [
     "nav", "footer", "#toc", ".link-popup", "script", "style",
     ".page-meta-footer", ".metadata", "[data-pagefind-ignore]",
@@ -204,8 +218,7 @@ def main() -> int:
         if neighbours:
             similar[page["url"]] = neighbours
 
-    SIMILAR_OUT.parent.mkdir(parents=True, exist_ok=True)
+    atomic_write_text(SIMILAR_OUT, json.dumps(similar, ensure_ascii=False, indent=2))
-    SIMILAR_OUT.write_text(json.dumps(similar, ensure_ascii=False, indent=2))
     print(f"embed.py: wrote {len(similar)} similar-links entries")
 
     # --- Semantic index (paragraph level) ---
@@ -221,12 +234,12 @@ def main() -> int:
         batch_size=64,
     ).astype(np.float32)
 
-    SEMANTIC_BIN.write_bytes(para_vecs.tobytes())
+    atomic_write_bytes(SEMANTIC_BIN, para_vecs.tobytes())
 
     meta = [{"url": p["url"], "title": p["title"],
              "heading": p["heading"], "excerpt": p["excerpt"]}
             for p in paragraphs]
-    SEMANTIC_META.write_text(json.dumps(meta, ensure_ascii=False))
+    atomic_write_text(SEMANTIC_META, json.dumps(meta, ensure_ascii=False))
 
     print(f"embed.py: wrote {len(paragraphs)} paragraphs to semantic index "
           f"({SEMANTIC_BIN.stat().st_size // 1024} KB)")

tools/sign-site.sh

@@ -40,15 +40,51 @@ if ! GNUPGHOME="$GNUPGHOME" gpg \
 fi
 echo "sign-site: pre-flight OK — signing $SITE_DIR..." >&2
 
-find "$SITE_DIR" -name "*.html" -print0 | xargs -0 -I {} -P $(nproc) \
+# Sign sequentially through a single gpg-agent: parallel signing causes
-    gpg --homedir "$GNUPGHOME" \
+# pinentry/IPC races where individual signs fail silently while xargs
-        --batch \
+# still exits 0. Atomic write via .tmp + mv avoids leaving a truncated
-        --yes \
+# .sig if the script is interrupted mid-write.
-        --detach-sign \
+sign_one() {
-        --armor \
+    local html="$1"
-        --local-user "$SIGNING_KEY" \
+    local sig="${html}.sig"
-        --output "{}.sig" \
+    local tmp="${sig}.tmp"
-        "{}"
+    if ! gpg --homedir "$GNUPGHOME" \
+             --batch \
+             --yes \
+             --detach-sign \
+             --armor \
+             --local-user "$SIGNING_KEY" \
+             --output "$tmp" \
+             "$html"; then
+        rm -f "$tmp"
+        echo "sign-site: FAILED to sign $html" >&2
+        return 1
+    fi
+    mv -f "$tmp" "$sig"
+}
+
+count=0
+while IFS= read -r -d '' html; do
+    sign_one "$html"
+    count=$((count + 1))
+done < <(find "$SITE_DIR" -name "*.html" -print0)
+
+# Post-sign manifest verification: every .html must have a non-empty
+# matching .sig. This catches any per-file failure that slipped through
+# (set -e bails on first failure inside the loop, but a manual --output
+# write to a directory containing a stale .sig from a prior run could
+# look "successful" otherwise).
+missing=0
+while IFS= read -r -d '' html; do
+    if [ ! -s "${html}.sig" ]; then
+        echo "sign-site: missing/empty signature for $html" >&2
+        missing=$((missing + 1))
+    fi
+done < <(find "$SITE_DIR" -name "*.html" -print0)
+
+if [ "$missing" -ne 0 ]; then
+    echo "sign-site: $missing HTML files lack signatures — aborting" >&2
+    exit 1
+fi
 
-count=$(find "$SITE_DIR" -name "*.html" -printf '.' | wc -c)
 echo "Signed $count HTML files in $SITE_DIR."