Detect verification-code signups (e.g. "your verification code is
371404") whose only link is a mailto. These cleared the keyword
threshold but were dropped because the detector required an http(s)
candidate link. A code path now raises the flag/badge/banner when a
verification keyword sits next to an OTP-style code; the code is never
extracted or surfaced.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The weak link-signal vocabulary was English-only, so a genuine double
opt-in whose confirm button reads "Je m'inscris…" over an opaque tracking
redirect scored 0 on every link and was missed. Make the weak vocab
multilingual (FR/DE/ES) to match the confirmation keywords.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
DuckDuckGo serves hi-res PNG favicons that legitimately exceed the old
100 KB cap, causing them to be rejected and negatively cached.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Surface each feed's email count on its Emails button and a "Last email …"
freshness line under the title, in both dashboard views. The values are
projected into feeds:list (kept to a single KV read) via the Feed aggregate,
so toListItemDTO now maps the whole aggregate through its intention-revealing
accessors instead of threading scalar projections. Also fixes long titles
overflowing into the Feed ID column in the table view.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Readers like FreshRSS discover the hub from <atom:link rel="hub"> in the
feed XML, not the HTTP Link header. Without it they never subscribe and
only refresh on cache expiry (~30 min) instead of receiving an instant
push when a new email arrives.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
linkedom escapes & in text nodes but not in attribute values, so URLs
with query strings (?a=1&b=2) serialized with bare ampersands. Valid XML
inside the feed CDATA, but the W3C validator parses the embedded HTML and
warns "Named entity expected. Got none." on <description>/<content:encoded>
(RSS) and <summary>/<content> (Atom). Escape every & not already starting
a valid entity; covers all three formats via processEmailContent.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
A failed favicon lookup was cached for a full week (same TTL as a
success), so a transient miss (e.g. the icon not yet indexed upstream)
blacklisted the domain for days. Cache negatives for 6 hours instead so
the next email retries.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Weak subscribe/subscription signals are now matched on the link href OR its
visible text (matched once, not additively), so a double opt-in email whose
button reads "Yes, subscribe me…" over an opaque tracking-redirect href is no
longer missed. Adds a regression test with anonymized fixture data.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Introduce CHANGELOG.md (Keep a Changelog) as the single source of release
notes, and scripts/release.sh (npm run release X.Y.Z) which promotes the
Unreleased section, commits the bare version as a real release commit, tags
it, and reopens the next -develop cycle. The Release workflow now verifies the
tagged commit's version equals the tag and publishes the CHANGELOG section as
the release notes instead of auto-generated commit lists.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Julien Herr