feat: syndicate blog posts to ATProto via standard.site lexicons#404
Merged
Conversation
Publish full blog posts as `site.standard.publication` + `site.standard.document` records on the PDS, and enhance the existing Bluesky POSSE syndication with `associatedRefs` for richer link cards (BLOG-b31d0c4d3b164931). - Add `publish-standard-site init <key>` / `sync [--all|--key]` CLI commands backed by a new `publications.toml` config. - Extend `BlogFrontMatter` with optional `atproto_uri` and a `publication` field (defaults to "blog"). These act as the idempotency keys for sync. - Extend `BlueskyClient` with `create_publication`, `put_publication`, `put_document`, `upload_blob`, and `create_blog_post`. `ExternalEmbed` now carries `associatedRefs` (skipped when empty). - Factor shared YAML frontmatter helpers into a new `commands::frontmatter` module; rewire the note publisher to use it. - Emit `<link rel="site.standard.document">` / `<link rel="site.standard.publication">` tags on rendered blog posts for verification, sourced from `publications.toml` baked in at compile time. - Extend the bluesky GH Actions workflow to run `sync --key blog` after the note publish step and commit any frontmatter updates. The standard.site publication record still needs to be bootstrapped against prod creds (`init blog`) before the workflow will succeed; until then the sync step errors out with an actionable message.
Make `publish-standard-site init` an idempotent no-op when the publication already has `at_uri` + `at_cid` cached in `publications.toml`, and wire it into the bluesky workflow as a pre-sync step. The first deploy after merge creates the publication record on the PDS and commits the populated config; every subsequent run short-circuits without touching the network. Removes the manual `init blog` bootstrap step from the merge checklist — the workflow now handles it. A new `--force` flag preserves the old behavior (re-`putRecord` the publication) for when config fields like `title`/`description` need to be pushed to the PDS.
- standard_site init now fetches the publication's branded OG card as a
PNG through imgproxy (`/og/publication/{key}.svg` → imgproxy 1200×630
PNG → upload as blob) instead of reading a static `opengraph.png`
from disk. New `/og/publication/{key}` SVG route on the server owns
the rendering; the CLI just wraps it in the same imgproxy URL format
used by per-post cards. `publications.toml` loses the `cover_image`
field; `PublicationConfig` no longer exposes it.
- Sequencing: the bluesky workflow `workflow_run`-chains off Fly
Deploy, so on merge the new SVG route is live before init runs.
- Bump review-app memory 512MB → 2048MB and add the env vars currently
missing from review-app secrets (boot-time required by
`AppState::from_env`): `ANTHROPIC_API_KEY`, `LINEAR_CLIENT_ID`,
`LINEAR_CLIENT_SECRET`, `LINEAR_WEBHOOK_SECRET`. Without these the
app crashes during `*Config::from_env()` before serving any request.
- Wire `IMGPROXY_URL` secret into the bluesky workflow init step so
init can rasterize the SVG card.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Boot crashes were the env-var gap (LINEAR_*, ANTHROPIC_API_KEY), not OOM. Keep memory at 512MB; revisit only if we see actual OOM signal. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The `2026-05-29` cutoff prevented historical posts from syndicating to standard.site. Drop it so the first post-merge run picks up the full back-catalog; subsequent runs stay cheap via the existing `atproto_uri` idempotency check in `classify_blog_post`. Per-post `published_at` is read from frontmatter `date`, so historical posts keep their original publication date on the PDS. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
`post_get` was calling `.unwrap()` on `fetch_thread`, which panics the entire page when Bluesky is down or rate-limits us. Mirror the notes handler: log the warning, render the page without the comments block. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Reintroduce the date cutoff (`2026-05-29`), but use it solely to gate Bluesky post creation — every post still gets a `site.standard.document` record on first deploy. Historical posts (before cutoff) syndicate to standard.site only, recent posts continue to get both. Avoids spamming the Bluesky feed with backdated posts on backfill. Idempotency: - Historical Both (no atproto_uri, no bsky_url): write atproto_uri only - Historical BskyOnly (atproto_uri set, bsky_url unset): terminal state, skip without calling put_document - Recent posts: unchanged — Both writes both, BskyOnly writes bsky_url Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Drives both the publication-level OG card text (currently looks sparse with just "coreyja") and the standard.site publication record's title field. Matching the displayed name to the actual domain reads better on the card and avoids name/URL drift in standard.site clients. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The publication card was looking sparse — big logo, small tag pill, one short title line, then a lot of empty space. Re-purpose the per-post-card date slot to render the publication description from \`publications.toml\` instead. For \"blog\", that surfaces \"Personal blog: Rust, side projects, Battlesnake, AI agents.\" under the title. Description is truncated to 75 chars (word-boundary + ellipsis) so future longer descriptions don't overflow the card width at Quicksand 28px. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Shift the tag pill (and everything below it) down 25-30px so the logo has more breathing room above the tag. Affects per-post cards too — the title and date moved with it to preserve their relative spacing to the tag pill. Date baseline now sits at y=595 (35px from the canvas bottom of 630), still comfortably inside the safe area. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds a \`subtitle\` frontmatter field to blog posts. When set, it renders in the bottom-left slot of the OG card (where the date used to be) and the date moves to a new bottom-right slot. When unset, the layout matches the previous behavior: date in the bottom-left, no bottom-right text. Publication cards continue to use the bottom-left slot for the publication description; their bottom-right slot is always stripped. Subtitle uses the same 75-char word-boundary truncation as the publication description so longer taglines don't overflow. Wires subtitle in for blog post (\`og_post_svg\`) and weekly newsletter (\`og_weekly_svg\`) cards; notes and podcast cards pass None for now (their frontmatter types don't carry a subtitle yet). Adds a subtitle to the \"Look Ma' no AI\" post as the first example. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ests Caught by workspace-wide clippy on CI (\`cargo clippy --all-targets\`); I was only running clippy on the server crate locally and missed the two literals in posts/src/blog.rs's own tests. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Tag pill bottom was at y=404, title baseline at y=455 — only ~5px between pill bottom and the visual top of the 64px title letters. Shift title down 10px (title line 1 → y=465, line 2 → y=535). Bottom slot at y=595 stays put; title-to-bottom gap tightens by 10px which still leaves comfortable room. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
coreyja
approved these changes
Jun 5, 2026
This was referenced Jun 5, 2026
coreyja
pushed a commit
that referenced
this pull request
Jun 6, 2026
## Summary The `Publish Newsletter` workflow has been failing on every push to `main` since 2026-06-05 13:42 UTC. The failure is **not** in the publish step — that succeeds and creates the Buttondown draft. The failure is the trailing `Commit buttondown_id updates` step: the default `GITHUB_TOKEN` isn't a bypass actor on the main-branch ruleset, so `git push` is rejected. Net effect: every push to main yesterday (every `Sync syndication state` commit) re-published ep3 as a fresh draft, failed to commit the id back, and the next push did it again. Ep 3 (`blog/weekly/20260407/index.md`) accumulated ~50 duplicate Buttondown drafts. ## Fix Mirror the GitHub-App-token pattern already used by `.github/workflows/bluesky.yml`: - Mint a token via `actions/create-github-app-token` (the App is the configured ruleset bypass actor). - Hand it to `actions/checkout` so subsequent git ops use it. - Check out `ref: main` rather than the trigger SHA so a queued run sees the freshest frontmatter. - `git pull --rebase origin main` before push to absorb any commits landed during the run. - Add `concurrency: newsletter-publish` (`cancel-in-progress: false`) so two pushes can't race the same unpublished file through publish and double-create the draft. ## Tests Two regression tests on `commands::buttondown::parse_frontmatter` covering the new `subtitle` / `atproto_uri` / `atproto_pub_cid` / `publication` fields from PRs #404-#408 — confirming serde is not the regression. All 25 buttondown tests pass; clippy clean (`-D warnings`). ## Followup not in this PR ~50 duplicate ep3 Buttondown drafts to clean up manually. Once this PR merges, the next push triggers the workflow, publishes ep3 one more time (51st), and the commit-back finally lands the `buttondown_id` so subsequent runs short-circuit. ## Test plan - [ ] Merge PR - [ ] Next push to main turns workflow green - [ ] `blog/weekly/20260407/index.md` ends up with `buttondown_id:` via auto-commit - [ ] Manually clean up duplicate Buttondown drafts (keep one canonical, ~50 dupes) 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Review Brief
What changed and why
This PR ships full blog posts to the ATProto PDS as
site.standard.publication+site.standard.documentrecords and enhances the Bluesky POSSE syndication so each bsky post carriesassociatedRefspointing at those records for richer link cards. Two new CLI commands (publish-standard-site init <key>andsync --key <key>) are added and wired into the GH Actions deploy workflow so syndication happens automatically after each Fly Deploy.Architecture decisions and trade-offs
putRecordis idempotent: re-syncing a post overwrites the same rkey rather than creating duplicates. Newsletter slugs (weekly/20230713/index.md→weekly-20230713) are handled by joining parent path segments with-.publications.tomlis the source of truth for publication config and cachesat_uri/at_cidafterinit.toml::to_string_prettystrips comments on round-trip, so the file must remain comment-free.include_str!bakespublications.tomlinto the binary at compile time for the blog page<link>tags. A parse error in the TOML surfaces as a.expect()panic on first HTTP request rather than a build failure. Missingat_uridoes not panic — the link tag simply isn't emitted.<link>tags lag one deploy: blog content isinclude_dir!'d at compile time, so the tags only appear after the next Fly Deploy following the frontmatter sync commit.putRecordalways bumpsupdatedAteven if post content is unchanged (accepted tradeoff for simplicity in theBskyOnlyrecovery path).init --forcechanges the publication's CID; all existing document records on the PDS embed the old CID. Recovery: re-runsync --key blogto re-putRecordall documents with the fresh CID.initis idempotent — it short-circuits whenat_uri+at_cidare already cached inpublications.toml. The workflow runsinit blogon every deploy; the first run creates the publication record and commits the populated config, every subsequent run is a no-op. No manual pre-merge step required.Risk assessment
<link>tags in<head>); Bluesky publish GHA workflow (new init+sync steps that can fail the run);BlogFrontMatterserialization (two new optional fields);ExternalEmbedstruct (newassociated_refsfield — both literal constructions updated).site.standard.*lexicon carries an explicit instability notice; new required fields in a future schema version would breakputRecord.BlogFrontMatterfields are optional/defaulted,OpenGraph.head_linksdefaults to empty, and the GHA sync step failure is surfaced independently from the note publish step. Revertingbluesky.ymland removingpublish-standard-sitefrom the binary restores prior behavior completely.Summary
Implements BLOG-b31d0c4d3b164931: publish full blog posts as
site.standard.publication+site.standard.documentrecords on the PDS, and enhance the existing Bluesky POSSE syndication withassociatedRefsfor richer link cards.publish-standard-site init <key> [--force]andpublish-standard-site sync [--all|--key <k>]initis idempotent: no-op whenat_uri+at_cidalready cached. Use--forceto re-upload config changes (title/description/cover).publications.toml(cachedat_uri+at_cidafter first init)BlogFrontMatterwith optionalatproto_uriandpublication(defaults to"blog")BlueskyClientgainscreate_publication,put_publication,put_document,upload_blob,create_blog_post;ExternalEmbedgetsassociatedRefsserver/src/commands/frontmatter.rs; note publisher now uses them<link rel="site.standard.document">/<link rel="site.standard.publication">head tags for verificationbluesky.ymlworkflow runsinit blog(auto-bootstrap, idempotent) followed bysync --key blogafter the existing note publish stepTest plan
cargo fmt --checkcleancargo clippy --all-targets --all-features --workspace --tests -- -D warningscleanall_publishable_blog_posts_fit_within_bsky_post_limitcovers allblog/**/index.mdrecursivelyall_blog_posts_have_unique_rkeysensures no two posts collide on the PDSpublications.toml; subsequent deploys log "already initialized" and skipCloses BLOG-b31d0c4d3b164931.