This repository ships about 1,300 free tools and a stream of research write-ups. A dry run on 20 October 2026 found 189 new items in its seven-day window: 184 tools and 5 posts. The owner asked for one tool that publishes to every platform where people's attention goes and measures which platforms work. This post covers what that tool does, what each platform actually permits, and what is still unknown.
Who did this. AI coding agents (Claude) working under the owner's direction, in one session. Nothing below has been posted to a live account yet, and no performance numbers exist. The interesting part is what the platforms permit.
One item, eighteen shapes
platform/distribution/hub.mjs takes each new tool or post and writes one version per platform. Each version fits the platform's limits:
- 280 characters on X, 300 graphemes on Bluesky and 500 on Threads;
- a caption without a clickable link on Instagram;
- a 30-second script on TikTok and YouTube Shorts;
- a "Show HN" title on Hacker News;
- a weekly digest by email.
Every version uses only the item's own title and description, so it cannot claim anything the page does not. Each version links back with a per-platform tag (?ref=threads, ?ref=x, ...). The site's counter adds one per platform per day when someone follows that link, and records nothing about who it was.
What the platforms permit
We sorted the 18 platforms by one rule from the repository's distribution spec: automate only what is free and within the platform's terms.
| Mode | Count | Platforms |
|---|---|---|
| Publish automatically (off until the owner switches it on) | 4 | Threads, Facebook Page, Instagram, email newsletter |
| Already automated by existing scripts | 4 | Bluesky, Mastodon, dev.to, Hashnode |
| Drafts for a person to post | 8 | X, LinkedIn, Substack, Medium, TikTok, YouTube, Reddit, Hacker News |
| Nothing to post (they find the site) | 2 | Search engines, AI assistants |
The eight draft-only platforms are drafts for different reasons:
- X made pay-per-use the default for its API on 6 February 2026 and closed the free tier to new developers. Third-party trackers put a post containing a link at about $0.20 each. Our rule is that distribution costs nothing, so X gets drafts.
- TikTok limits unaudited apps to private posts (
SELF_ONLY) from private accounts, and every post needs a video. - YouTube also needs a video for every post, and uploads from unverified API projects stay private.
- Substack has no publishing API. Its 2026 developer API looks up profiles. The unofficial endpoints some tools use may break its terms.
- Medium no longer issues integration tokens. Its "Import a story" keeps the canonical link, so a person can still cross-post.
- LinkedIn lets a member post by API. Reading those posts back, which we need in order to avoid posting twice, requires partner-only access.
- Reddit and Hacker News forbid automated submissions.
Social media now splits into platforms you can automate cheaply (Meta's Threads, Facebook and Instagram APIs, Bluesky, Mastodon) and platforms where automation costs money or an audit. For a free project, the second group is where a person still has to post by hand.
Measuring what works, honestly
The scoreboard (hub.mjs score) combines two kinds of evidence and keeps them apart:
- Engagement per post, where the platform reports it:
- Bluesky and Mastodon expose likes, reposts and replies publicly;
- Threads has an insights API with views;
- Facebook and Instagram return reactions and comments;
- dev.to returns page views.
- Arrivals: page loads on our site that came through each platform's tagged link. This is the only number that shows a post sent someone to a tool.
Two safeguards keep early numbers from misleading:
- Shrinkage. A platform's rate per post is pulled toward the average by five pseudo-posts, so two lucky posts cannot outrank fifty steady ones.
- A floor. Nothing with fewer than five posts is ranked.
Short posts also rotate between a few honest wordings ("New free tool: …", "Free and open source: …"). The pick is a hash of the item and the platform, so a re-run writes the same text. Each wording starts with a fixed marker, so the scoreboard can tell which one a post used and compare them within each platform.
Current state: no readings. Nothing has posted yet, and the site's counter is not deployed, so every arrival count reads "unmeasured", never zero.
The same idea, for anyone building a brand
The second tool, Interview Content Studio, does this for a person or a business rather than for this repository:
- It interviews you about your brand or topic. You can keep several interviews.
- It sizes a posting plan for 14 platforms to the hours you actually have.
- It lays the plan out as a four-week calendar.
- It drafts every slot in that platform's format.
Without an API key, drafts are assembled only from your own answers. With your own key (Anthropic, OpenAI, Gemini, or a model on your own computer), drafts are rewritten by AI and then checked: a draft that introduces a number not in your interview is thrown away. It exports CSV for a scheduler and posts nothing itself. Its posting amounts are starting assumptions, and the page says so.
Limits
- No adapter has been exercised against a live account; the request shapes come from public documentation read on 20 October 2026.
- No engagement or arrival numbers exist yet, so this post makes no claim about which platform works best. That is the next post, once there is data.
- Platform rules change quickly. The tool records the date each fact was checked (
checked:inhub/platforms.mjs).
The full report and the next steps are in portfolio/distribution/hub-report-2026-10-20.md.