Not famous. Not abandoned. Just building.
A static site that surfaces open source projects flying under the radar that are still shipping: newborns, solo builders, projects that came back after a long pause. Ranked by activity relative to stars and watchers.
Live: https://silentstars.dev
| Home | Project page | Weekly edition |
|---|---|---|
![]() |
![]() |
![]() |
A directory of open source projects that are actively maintained but under-recognized. Every night the pipeline scores each tracked project on two axes — healthScore (is it alive right now?) and undervaluedScore (signal relative to reach) — classifies it into one of 7 vital states, and tags it with behavioral signals (solo_builder, hidden_gem, fork_magnet, funded, etc.) that drive the home page sections. One project is highlighted daily on Bluesky (@silentstars.dev), and every Monday a weekly edition rounds up the projects that moved — rising, shipping, lighting up — and goes out as a Bluesky thread. About 420 projects are listed as of November 2026.
- Astro + TypeScript (content collections)
- GitHub Actions (nightly discover → collect → build → deploy)
- GitHub Pages — zero cost, zero backend
scripts/discover.ts queries the GitHub Search API each night and writes candidates
to data/discovered.json. scripts/collect.ts then merges all sources, deduplicates,
and drops any discovered repo whose healthScore falls below the threshold.
Configure discovery in data/discovery.config.json:
{
"starsRange": { "min": 5, "max": 500 },
"pushedWithinDays": 90,
"maxCandidatesPerNight": 250,
"vitalityThreshold": 40,
"excludeRepos": ["capuz/silentstars"]
}| Field | Description |
|---|---|
starsRange |
Avoids empty repos (0–4) and larger ones (>500) |
pushedWithinDays |
Only considers repos pushed to recently |
maxCandidatesPerNight |
Hard cap — prevents exhausting the 30 req/min rate limit |
vitalityThreshold |
Discovered repos with healthScore below this won't appear in the directory |
excludeRepos |
Repos never picked up by discovery (e.g. this one) |
An array of { repo, until } entries:
[
{ "repo": "Code-Society-Lab/matrixpy", "until": null },
{ "repo": "carlos-menezes/caravan", "until": "2026-09-27" }
]until: null— permanent curation.until: "<date>"— time-boxed; the entry is pruned once it expires.
This is an inclusion mechanism, not a classification one: a promoted repo still
gets its status computed normally by computeStatus() (see Vital states).
Promotion only bypasses the healthScore threshold that would otherwise exclude it — it
has no relationship to the watched status.
Open an issue using the Submit a project template. process-submission.yml validates
the repo against the GitHub API (public, not archived, 500 stars or fewer, at least
one commit in the last 90 days), adds it to data/promoted.json with a 30-day window, and
runs the full collect → build → deploy → Bluesky post pipeline. The submitter gets an
acceptance email (see Email notifications), and Claude posts a
short take on the project as an issue comment (see below).
Brand-new repos with no outside signal yet (0 stars and 0 forks) aren't rejected — they're
parked on a 90-day standby in data/standby.json, the submitter is told why, and the issue
is closed. standby-recheck.yml checks them daily: once a repo gets a star or fork from
someone other than its owner, the submission is reprocessed and published; entries still
silent after 90 days are dropped.
Accepted submissions go live on the site right away, but their Bluesky posts go through a
queue in data/bsky-queue.json (scripts/bsky-queue.ts): at most 3 community picks
per UTC day, oldest first. On a quiet day the queue is empty and the post goes out a few
minutes after the deploy. Past the cap, or when a deploy times out, the issue is labelled
bsky-queued with its place in line, and bsky-queue.yml (3 times a day) posts it,
edits the issue comment with the link, emails the submitter and closes the issue.
- Create a GitHub personal access token (read-only, public repos scope is enough).
- Export it:
export GITHUB_TOKEN=ghp_your_token_here - Run the pipeline:
npm run pipeline # discover + collect in sequence # or step by step: npm run discover # writes data/discovered.json npm run collect # merges sources → data/latest.json + src/content/projects/
A few spots in the pipeline lean on Claude Code CLI (claude -p) for text generation.
All of them are optional and fail gracefully without a token — nothing in the pipeline
depends on Claude Code being available.
- The daily post's hook —
scripts/post-pitch.tsprefers the GitHub description or README first line for the "what it does" line of the Bluesky post. Only when that's too weak (status text, a bare pointer, too short) does it fall back toscripts/claude-cli.tsto generate one. - Translation —
scripts/collect-i18n.tsdetects non-English descriptions/READMEs and translates them via Claude Code, caching results indata/translations.jsonkeyed by repo and content hash so an unchanged project never costs a repeat call. Translated pages say so and link to the original on GitHub. - Weekly notes —
scripts/weekly-notes.tswrites the one-line note under each project in the weekly edition from its commits and releases.
scripts/notify-owners.ts— lets the owner of an auto-discovered project know it was featured (only if their GitHub profile has a public email); idempotent viadata/posted.json, so a repo is never emailed twice.scripts/send-email.ts+scripts/email/— sends "accepted" / "already listed" emails to submitters via Resend.
Requires RESEND_API_KEY and NOTIFY_FROM_EMAIL secrets.
npm install
npm run dev # http://localhost:4321/
npm run build # static output in dist/
npm run preview # serve dist/ locally| Weight | Dimension | Notes |
|---|---|---|
| 35% | Recency | Linear decay over 90d from last commit |
| 25% | Commit cadence | Consistency of commit intervals |
| 20% | Issue health | closedIssues ÷ totalIssues |
| 20% | PR health | mergedPRs ÷ totalPRs |
Bands: 🟢 Healthy 80–100 · 🟡 Stable 60–79 · 🟠 Quiet 40–59 · 🔴 At Risk <40
Repos with high signal and low reach rank highest. The directory is sorted by this.
signal = 0.25 × commitVelocity (commits in 90d, 30=max)
+ 0.20 × contributorWork (uniqueAuthors × commits90d / 100)
+ 0.20 × issueResolution (closedIssues ÷ total)
+ 0.20 × forkRatio (forks ÷ stars)
+ 0.10 × releaseCadence (releases in 90d, 3=max)
+ ageBonus (up to 0.3 if repo > 6 months old, linear up to 24m)
+ homepageBonus (0.05 if repo has a homepage URL)
reach = log10(stars + watchers + 10)
score = round((signal / reach) × 100), capped at 100
| State | Meaning |
|---|---|
thriving |
Commit in the last 14 days |
quiet |
None of the other states apply |
at_risk |
2–12 months with no commits + open issues/PRs |
newborn |
Created < 2 months ago, README + license + commit in the last 14 days |
revived |
Gap of > 6 months between commits; latest commit in the last 30 days |
watched |
Small repo with an unusually high watcher-to-star ratio (watchers ÷ stars ≥ 2, watchers ≥ 5, stars ≤ 150) — people are following closely before it has star traction |
archived |
Officially archived or no commits in 18+ months — hidden from directory |
watched is a computed status, not manual curation — don't confuse it with promotion via
data/promoted.json above, which is a separate, orthogonal mechanism (it controls
inclusion, not status).
| Workflow | Trigger | What it does |
|---|---|---|
collect.yml |
Nightly cron + push to main |
Core pipeline: collect → deploy → post → redeploy → notify-owner (see Deployment) |
process-submission.yml |
Issue opened | Validates a community submission, adds it to data/promoted.json, runs the full pipeline, notifies the submitter |
post-manual.yml |
Manual (workflow_dispatch) |
Re-post a specific project or the daily pick to Bluesky, with a dry-run option |
retry-nightly.yml |
On completion of collect.yml |
Reruns the nightly job if GitHub never assigned it a hosted runner |
post-weekly.yml |
Mondays 13:17 UTC | Posts the latest closed /weekly/ edition to Bluesky as a thread |
standby-recheck.yml |
Daily | Drops expired standby submissions and reprocesses those that gained outside signal |
A handful of other workflow_dispatch-only workflows (OG previews, test emails, Bluesky profile updates, recovery) are maintenance tools.
collect.yml ("Nightly collect + deploy") runs on a nightly cron and on push to main,
with five jobs in sequence:
collect— discover + collect, computes scores, commitsdata/latest.json+ content files + prunedpromoted.json+data/translations.json.deploy—astro build, screenshots OG cards via Playwright (scripts/og-screenshot.ts), deploys to GitHub Pages.post— posts the day's highlight to Bluesky.redeploy— rebuilds and redeploys so the project's page can show the Bluesky post URL from step 3.notify-owner— emails the featured project's owner, for organically-discovered picks only (submitters are already notified separately).
Use post-manual.yml to trigger a manual (re)post outside the nightly schedule.
In the repo settings, set Pages source to GitHub Actions.
Licensed under the Apache License 2.0.


