Backfeed only read first-level replies (Threads /replies, GoToSocial
direct replies, Bluesky depth 1) and skipped the owner's own replies, so
conversations under crossposts lost everything after the first reply.
- Threads /conversation, all GoToSocial descendants, Bluesky depth 10
- backfeed.thread_members keeps others' replies to our copy and any reply
(ours included) to a kept reply; own replies are stored as notified
- Migration 026: webmentions.parent_remote_id; replies render nested as
comments of comments
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mastodon turns GIFs into silent looping MP4s (gifv); only the first-frame
preview was kept. The MP4 is now saved locally and played muted and
looping with the still as poster. Animated GIFs from other sites are saved
as animated WebP. Migration 025 adds reply_contexts.video and has
fediverse contexts with an image fetched again once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Native replies on the same network: GoToSocial in_reply_to_id (resolved
by our instance) with the author's @mention; Bluesky reply root/parent
- Replies to our own notes (permalink or any crosspost URL) thread under
that note's crosspost on every network, Threads included
- Otherwise "↩ Replying to <author>" and the link; on Bluesky the name is
the link. compose_text keeps the line whole when shortening
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replies to Mastodon posts showed only the server's domain: the page has no
h-entry, so nothing was found and the host stood in for the author.
- Context sources in order: Bluesky AppView API, microformats, the
origin's Mastodon API or this site's GoToSocial instance, OpenGraph
- Migration 024: author_handle, photo, photo_alt, source; rows without a
source are fetched again once by the webmention worker
- Note page: a card with avatar, name, handle, text, image and date;
home/tag/likes: "In reply to <author>" with a short quote
- Avatars and images saved locally (CSP allows only this site's images);
avatar upkeep now covers reply contexts so their avatars aren't removed
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Links to the site itself, and to self-hosted apps like Gitea, timed out
from inside the container and sat in the admin Log. They are now skipped
when queueing; already queued or failed ones are dropped at worker start.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Crossposting runs in a background worker; url_for there raised and every
crosspost failed. post_url now joins SITE_URL and post_path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Articles (notes with a title) move from /note/<slug> to /article/<slug>;
the other prefix 301-redirects. New articles take their slug from the
title (mp-slug still wins; notes keep timestamp slugs). A Micropub update
that adds or removes a title answers 201 with the new Location.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Content sent as both html and text is stored as the HTML; the text is
usually a plain fallback, so images, links and headings were lost.
- Images embedded from other sites are copied into the media library
(SSRF-checked fetcher, best effort, max 20) and the text points at the
copies, on create and on content replace.
- HTML content is sanitized as is instead of going through Markdown, which
added stray <br>s and empty paragraphs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Articles as the IndieWeb defines them: a name separate from the content.
- Migration 023 adds notes.title and notes.summary. A one-time Python data
step (migrations.DATA_MIGRATIONS) moves each note's '# Heading' first line
into title, rewrites the file and hash, empties FTS for a rebuild, and puts
files back if the migration fails. Trashed notes included.
- Micropub: name -> title, summary stored when sent with content, summary
updatable, q=source returns both. MCP tools take a summary.
- Feeds: titles only for articles (RSS/JSON none, Atom empty); Atom and
JSON Feed summaries; the title is no longer repeated in the content.
- Pages: p-name/p-summary on the article page. Homepage and tag pages show
articles as title, summary and "Read more" without images (hidden u-photo
keeps the post type for parsers).
- Crossposts: articles go out as title, summary, link and hashtags, no
images; the Bluesky card is the article. Notes unchanged.
- Editor: Title and Summary fields; an article may have no body.
- Fix: a line starting with a hashtag ('#indieweb ...') was rendered as a
heading and crossposted without its '#' (HashtagLineExtension).
- Fix: excerpts showed raw Markdown; they are plain text now.
Back up the database and the data folder before deploying: the migration
rewrites note files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Architecture, README, standards, CLAUDE.md, settings reference, indexes,
backlog, release plans and CHANGELOG checked against the code at v2.14.0:
backfeed, notifications, counts line and web actions documented; test
policy (changed area locally, full suite in CI); stale versions, statuses
and links fixed; legacy and superseded docs moved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Likes, reposts, quotes and replies to a note's GoToSocial, Bluesky and
Threads copies are read back from the silos and stored as webmentions
(ADR-073): age-based schedule to 90 days, pruning of withdrawn responses,
#nobridge/#nobot opt-outs, admin backfill and Check now, Threads counts,
a Quotes group, local avatar copies, and delete tombstones.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The in-memory metrics system covered one gunicorn process at a time, was
lost on restart and was never used. Remove /admin/metrics-dashboard,
/admin/metrics, starpunk/monitoring, the HTTP/query timing, memory monitor,
feed counters, METRICS_* settings and psutil. Keep /health, /admin/health
(without its metrics block), pool stats and the feed cache. ADR-072.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fetch_source returns (content, content_type, final_url) since v2.7.0, but
fetch_reply_context unpacked two values, so every new reply/like/repost
context was stored as an error. Parse with the final URL so relative links
resolve. Adds a test that runs the real fetch path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- architecture/: overview, technology stack, auth (OIDC + built-in
IndieAuth), syndication (POSSE) rewritten; new database.md,
deployment.md and indieweb-protocols.md; legacy stack marked historical
- decisions/: reconstructed ADR-060 and ADR-063; 31 status corrections
(superseded auth ADRs, proposed -> accepted); INDEX rebuilt
- design/: reconstructed the v1.6.0 documents the webmention code cites;
reports for 18 v2 releases without a design folder; INDEX rebuilt
- projectplan/: BACKLOG rewritten (shipped items removed, open issues from
the reviews added); RELEASE.md for v2.0.0-v2.10.1; INDEX
- standards/: development setup (uv, tests, DEV_MODE), testing (CI test
stage), git and release flow, versioning, coding standards, cookies
- README, CLAUDE.md, .env.example (every setting the code reads, with
defaults) brought up to date
- 137 broken links repaired (old absolute paths, moved files); stale ADR
references in code comments corrected (comments only)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- mp-syndicate-to/mp-slug read from JSON properties; absent means no
crossposting
- Sanitize rendered HTML (nh3 allowlist)
- q=source/q=category need a posting scope; draft scope creates drafts;
private/unlisted visibility saved as drafts
- Photos that can't be fetched or found fail the request; remote photos
via the SSRF-checked, pinned fetcher
- JSON scalars wrapped, update delete values must be arrays
- bookmark-of stored; HTML content returned as {html} (migration 015)
- published stored as UTC, bad dates 400; malformed JSON 400
- Auth: case-insensitive Bearer, header+body 400, WWW-Authenticate
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
create_migrations_table() committed right after BEGIN IMMEDIATE, and
executescript() commits before running, so gunicorn workers applied
migrations concurrently; the v2.8.0 deploy failed to boot once with
'duplicate column name'. Run migration statements with conn.execute inside
the transaction and don't commit when creating the tracking table.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Responses as u-comment/u-like/u-repost/u-bookmark/u-mention h-cites
with author h-cards; source's own url and published (migration 014)
- dt-* and Micropub published in RFC 3339 UTC
- Title only for a real Markdown heading (hashtag notes stay notes)
- Ignore webmentions from a note's own syndicated copies
- Micropub/IndieAuth links independent of Webmention; rel=me/rel=author
- Representative h-card and authorship algorithm (starpunk/mf2_utils.py)
- Tag/likes h-feeds get url and author; previews get u-uid
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Connect to the address that passed the SSRF check (Host header and SNI
carry the hostname, so TLS still verifies it), closing the DNS rebinding
gap (spec 4.5). The webmention POST uses the same path.
- Treat a change in the responses a note shows as an update and re-send its
webmentions (spec 3.1.4), from validation and from admin moderation.
Unchanged re-validations send nothing, so there is no ping-pong.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sending: RFC 8288 Link header parsing, rel token matching and document
order for <link>/<a>, endpoint re-discovery on every send, re-send to
previous targets on update/delete/undelete, 410 Gone for deleted notes,
no redirect following on the POST, 429 retried.
Receiving: exact target match per media type, Accept header, final URL
for relative links, only published notes accepted, fragments ignored,
moderation kept on updates, transient errors keep valid mentions.
Security: every redirect hop SSRF-checked, all resolved addresses
checked, broader blocklist, total time and size limits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Normalise mf2 author values: u-photo with alt text is a dict in mf2py and
broke the database update for most real senders
- Pass str(response.url) to urljoin during endpoint discovery; the TypeError
marked every target as having no endpoint. Retry on discovery errors
- Hold a cross-process flock so only one gunicorn process runs the worker
- Keep author URL/photo only if http(s) (XSS via javascript: URLs)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The worker thread called the SSRF checks, which read current_app.config,
without an app context, so every incoming and outgoing webmention failed with
'Working outside of application context'. A teardown_appcontext hook also
stopped the worker whenever a request ended.
Pass the app to WebmentionWorker and run each poll in app.app_context(), like
SyndicationWorker, and drop the teardown hook.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Uncaptioned photos rendered alt="Image", which mf2 parsers report as
the photo's alt text. They now take the post's text (excerpt, up to 150
characters), falling back to "Image" only for text-free posts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Checked rendered pages with mf2py against W3C Post Type Discovery:
- Note pages put u-in-reply-to inside a bare h-cite, making it the
h-cite's property, so replies parsed as notes (or photo posts). The
reply context is now the entry's u-in-reply-to h-cite.
- Home page previews had no reply markup; they now show "In reply to"
with u-in-reply-to.
- Previews always emit e-content, so a text-free photo post can't get
an implied name.
- Text-free page titles follow PTD order (Repost, Like, Reply, Photo).
tests/test_post_types.py runs PTD on parsed note pages and the home page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A create that failed after the note was written (e.g. the same photo
twice) answered 500 but left the post published without photos.
Photos are now resolved first, and a failed note is removed.
- Pooled connections roll back unfinished writes, so one failure no
longer locks the database for the next ~10 s.
- The same photo twice in one post is shown once.
- Photo posts no longer need text (Micropub doesn't require content,
and q=config already advertised a photo-only post type). Updates may
clear the text while photos remain. Crossposts start with the link;
the page is titled "Photo".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Third crossposting target on the ADR-065 loop (ADR-069). Text posts get a
link preview, 1 image posts as IMAGE and 2-4 as a CAROUSEL (Meta fetches
them from /media/), with alt text and the first tag as topic_tag. Posts are
shortened to 500 characters, counting non-ASCII as UTF-8 bytes. Edits don't
follow (the API has no edit); unpublish and delete remove the post. Retried
creates check for an existing post first.
The 60-day token is refreshed daily from the worker loop and kept in
DATA_PATH/threads_token.json under a file lock; a new configured token
replaces it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Micropub `name` was dropped; store it as the note's '# Heading' first
line, which is how StarPunk titles work. q=source returned /notes/<slug>
(a 404) and a name for every note; it now returns the real permalink and
name only for titled notes, without the heading in content. Titled notes
no longer show their title twice (Note.body_html, excerpt).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Admin login now uses OpenID Connect (PKCE, nonce, ID token verified
against the provider JWKS) and only OIDC_ADMIN_SUB may log in. Micropub
tokens are issued and verified by StarPunk itself, with a consent screen
behind the OIDC session and a Connected Apps page to revoke them, so an
external IndieAuth server (Gondulf) is no longer needed.
Existing notes, media, tags, webmentions, author profile and sessions
are unchanged. Migration 011 adds new tables only.
BREAKING CHANGE: production requires OIDC_ISSUER, OIDC_CLIENT_ID,
OIDC_CLIENT_SECRET and OIDC_ADMIN_SUB. The ADMIN_ME page must point
Micropub apps at StarPunk's IndieAuth endpoints. See ADR-064.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
IndieAuth Authentication:
- Corrected W3C IndieAuth specification compliance
- Uses response_type=id for authentication-only flow per spec
- Discovers endpoints from user profile URL
- Removed hardcoded indielogin.com service
- DEPRECATED: INDIELOGIN_URL config (now auto-discovered)
Timestamp-Based Slugs (ADR-062):
- Default slugs now use YYYYMMDDHHMMSS format
- Unique collision handling with numeric suffix
Debug File Management:
- Controlled by DEBUG_SAVE_FAILED_UPLOADS config
- Auto-cleanup of files older than 7 days
- 100MB disk space protection
- Filename sanitization for security
Performance:
- N+1 query fix in feed generation
- Batch media loading for feed notes
Data Integrity:
- Atomic variant generation with temp files
- Database/filesystem consistency on failure
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
BREAKING: Removes INDIELOGIN_URL config - endpoints are now properly
discovered from user's profile URL as required by W3C IndieAuth spec.
- auth.py: Uses discover_endpoints() to find authorization_endpoint
- config.py: Deprecation warning for obsolete INDIELOGIN_URL setting
- auth_external.py: Relaxed validation (allows auth-only flows)
- tests: Updated to mock endpoint discovery
This fixes a regression where admin login was hardcoded to use
indielogin.com instead of respecting the user's declared endpoints.
Version: 1.5.0-hotfix.1
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Per v1.5.0 Phase 4:
- Generate variants to temp directory first
- Perform database inserts in transaction
- Move files to final location before commit
- Clean up temp files on any failure
- Add startup recovery for orphaned temp files
- All media operations now fully atomic
Changes:
- Modified generate_all_variants() to return file moves
- Modified save_media() to handle full atomic operation
- Add cleanup_orphaned_temp_files() for startup recovery
- Added 4 new tests for atomic behavior
- Fixed HEIC variant format detection
- Updated variant failure test for atomic behavior
Fixes:
- No orphaned files on database failures
- No orphaned DB records on file failures
- Startup recovery detects and cleans orphans
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Per v1.5.0 Phase 3: Fix N+1 query pattern in feed generation.
Implementation:
- Add get_media_for_notes() to starpunk/media.py for batch media loading
- Add get_tags_for_notes() to starpunk/tags.py for batch tag loading
- Update _get_cached_notes() in starpunk/routes/public.py to use batch loading
- Add comprehensive tests in tests/test_batch_loading.py
Performance improvement:
- Before: O(n) queries (1 query per note for media + 1 query per note for tags)
- After: O(1) queries (2 queries total: 1 for all media, 1 for all tags)
- Maintains same API behavior and output format
All tests passing: 920 passed in 360.79s
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Replaces content-based slug generation with timestamp format YYYYMMDDHHMMSS.
Simplifies slug generation and improves privacy by not exposing note content in URLs.
Changes:
- Add generate_timestamp_slug() to slug_utils.py
- Update notes.py to use timestamp slugs for default generation
- Sequential collision suffix (-1, -2) instead of random
- Custom slugs via mp-slug continue to work unchanged
- 892 tests passing (+18 new timestamp slug tests)
Per ADR-062 and v1.5.0 Phase 1 specification.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- ADR-062: Timestamp-based slug format (supersedes ADR-007)
- Updated v1.5.0 RELEASE.md with 6-phase plan
- Updated BACKLOG.md with deferred N+1 query locations
- Developer questions and architect responses for Phase 1
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add pillow-heif dependency for iPhone photo support
- Auto-convert HEIC to JPEG (browsers can't display HEIC)
- Graceful error if pillow-heif not installed
- Handles RGBA/P mode conversion to RGB
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Implements media upload logging per docs/design/v1.4.1/media-logging-design.md
Changes:
- Add logging to save_media() in starpunk/media.py:
* INFO: Successful uploads with file details
* WARNING: Validation/optimization/variant failures
* ERROR: Unexpected system errors
- Remove duplicate logging in Micropub media endpoint
- Add 5 comprehensive logging tests in TestMediaLogging class
- Bump version to 1.4.1
- Update CHANGELOG.md
All media upload operations now logged for debugging and observability.
Validation errors, optimization failures, and variant generation issues
are tracked at appropriate log levels. Original functionality unchanged.
Test results: 28/28 media tests pass, 5 new logging tests pass
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Complete Phase 5 of v1.4.0 "Media" release with comprehensive release
preparation.
CHANGELOG Updates:
- Added v1.4.0rc1 section with comprehensive release notes
- Documented all four phases:
* Large Image Support (50MB limit, tiered resize)
* Image Variants (thumb, small, medium, large)
* Micropub Media Endpoint (W3C compliant)
* Enhanced Feed Media (Media RSS, JSON Feed extension)
- Standards compliance documented (W3C Micropub, Media RSS 2.0)
- Storage impact and backwards compatibility notes
- Configuration options (STARPUNK_ABOUT_URL)
Version Bump:
- Updated starpunk/__init__.py to 1.4.0rc1
- Version info tuple includes rc1 designator
Implementation Report:
- Created docs/design/v1.4.0/2025-12-16-v140-implementation-complete.md
- Complete summary of all five phases
- Test results: 850 passing, 19 failing
- Known issues documented (flaky tests, feed cache)
- Acceptance criteria checklist
- Deployment and rollback procedures
- Release recommendation: Ready for RC testing
Test Results:
- Full test suite executed: 850 passed, 19 failed (98% pass rate)
- Core v1.4.0 functionality verified working
- Failures in known flaky tests and unrelated areas
- 5 minutes 58 seconds execution time
Release Status:
- All five phases complete
- Ready for release candidate testing
- Manual verification recommended before final release
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>