post-publisher

Input: an output directory from carousel-builder / reel-builder (or a single file, or just text). Output: a live post or a platform-side draft, plus a posted.json receipt.

This skill does NOT:

• Write the content — that is viral-text / essay-write / carousel-builder / reel-builder.
• Schedule anything. There is no queue and no cron. It publishes now, or it stages a draft you finish by hand.
• Post to a platform you have not connected. Credentials are yours; nothing is bundled.
• Guarantee reach, or touch analytics.

ROLE

Resolve a source directory → build one platform-neutral Post → preflight it against each target platform's real limits → show the user exactly what would go where → ask → publish → write a receipt.

Publication is irreversible. Dry-run is the default and --yes is what leaves it. Even with --yes, each platform is confirmed separately: approving an Instagram post is not approval to also post to X.

PIPELINE

1. Resolve the source. A directory is inspected in this order: final.mp4 → video; several images → carousel (sorted numerically, so slide-2 precedes slide-10); one image → image; nothing → text post. Override with --kind.

2. Resolve the caption. From --text, else --text-file, else captions.md in the source directory. The parser looks for a ## Main post / ## Основной пост heading and takes everything up to the next heading; with no such heading it takes the whole file minus headings. Always check the extracted text in the dry-run preview — captions.md is written by an agent, so its shape is a convention, not a contract.

3. Preflight, per platform. Character limits including hashtags, media count, file size, extension, missing alt text, and each platform's own rules (Telegram's 1024-char caption ceiling that does not apply to text posts; Instagram's 24h posting cap; TikTok's audit caveat). A block finding means nothing is sent; a warn means it goes but you should know.

4. Preview + confirm. Full text, character count, every file with its size and alt text, and whether it is going out live or as a draft.

5. Publish, then record. The receipt (posted.json) is written immediately after each platform succeeds — a crash mid-fan-out cannot cause a re-post of the platforms already done. A later run with identical content is refused unless --force.

PLATFORMS

Platform	Post kinds	Draft	The thing that will bite you
telegram	text, image, carousel, video	—	Caption on media caps at 1024 chars (text-only gets 4096). Video ≤50 MB. The only one testable end-to-end without OAuth — start here.
threads	text, image, carousel, video	container	500 chars. Text-only posts need no S3 — the easiest real connection.
instagram	image, carousel, video	container	Business/Creator account required. Media is fetched by URL, so S3 is mandatory. Video publishes as a Reel (≤300 MB, ≤15 min). 100 posts/24h.
tiktok	video, image, carousel	inbox	Direct publishing needs an audited app; without one, posts are silently forced to SELF_ONLY. Use --draft.
x	text, image, carousel, video	—	280 chars. 10,000/24h per app, 100/15min per user; what you may spend depends on the plan. Threads publish incrementally — a failure mid-chain leaves earlier posts live.
youtube	video	private	100 uploads/day on their own allocation, drafts included. Cleanest draft of the seven.
linkedin	text, image, carousel, video	—	3000 chars. Member posts work with w_member_social; company pages need the partner-gated Community Management API.

MODES

Source

• post-publisher <dir> — a carousel-builder / reel-builder output directory
• post-publisher <file> — a single image or video
• post-publisher --kind text --text "..." — no source at all

Targeting

• --platform a,b,c — comma-separated; each is confirmed separately
• --draft — stage instead of publish. A platform without drafts is skipped, not published live
• --kind text|image|carousel|video|article — override auto-detection

Content

• --text / --text-file — caption (overrides captions.md)
• --title — YouTube title / LinkedIn article title
• --hashtags "a,b,c" — kept separate from the body so preflight can count them
• --alt "..." — repeat once per media file, in order
• --link — attached where the platform supports it

Article syndication (--kind article)

• --canonical <url> — the important one. The URL of the original on your own domain
• --description — SEO excerpt / subtitle
• --series — dev.to series name, Hashnode series id
• --cover-url — cover image, already hosted
• --packets <dir> — write submission packets for the platforms with no API

Execution

• (nothing) — dry-run. Prints what would happen and stops
• --yes — leave dry-run. Still asks before each platform
• --force — publish despite a matching receipt in posted.json
• --publish-container <id> — publish a Meta container left by an earlier --draft
• --check --platform X — readiness: credentials present, token usable
• --list-platforms — everything, with what each is missing

Exit codes

Code	Meaning
0	Everything asked for happened, or was already done
1	At least one platform failed — blocked by preflight, missing credentials, or a platform error
2	Usage error: no --platform, unreadable source

A skip is not a failure. A platform skipped because it has no drafts, or because posted.json says this content already went out, leaves the code at 0 — in a mixed run (--platform telegram,instagram --draft) Instagram drafting successfully should not be reported as an error because Telegram was skipped. The summary line always states how many proceeded, were skipped and were blocked.

Accounts

• python3 -m common.runners.cli.auth --platform <name> — OAuth in a browser
• ... --paste-token — paste a token from the platform's own token tool (the supported route wherever loopback redirects are rejected)
• ... --status / --verify / --revoke <name>
• skills-keys accounts — the same list, from the credentials skill

REFERENCES (load on demand)

File	When to load
references/syndication.md	--kind article — why canonical decides whether syndication helps or hurts, which platforms have an API, and the order to publish in
references/oauth-setup.md	First — per-platform app registration, scopes, account type, which env vars go in ~/.skills.env
references/platform-limits.md	Step 3 — the limit matrix preflight enforces, with what to verify against live docs
references/browser-fallback.md	When the API path is closed: unaudited TikTok, personal Instagram, org LinkedIn pages
references/troubleshoot.md	A platform returned an error and the message is not self-explanatory
viral-text platform budgets	Writing to fit a platform before generating — character budgets and tone per network

EXAMPLES

See examples/before-after.md — three calibration runs: an 8-slide carousel to Instagram as a draft, a reel to TikTok's inbox, and a text post fanned out to Threads + Telegram + X.

CONSTRAINTS

• Dry-run is the default and must stay that way. Every other runner here is safely re-runnable because a wasted generation costs cents. A wasted publish costs an audience. --yes is the deliberate act.

• Confirm per platform, never in bulk. A single "yes" must not fan out to seven networks.

• --draft is a promise. If a platform cannot draft, skip it. Publishing live because the user asked for a draft is the worst possible reading of the flag.

• Never print a token. Masked in --status, absent from errors, absent from logs. ~/.skills-tokens.json is chmod 600 and gitignored.

• Preflight before bytes. Every check that can happen locally happens before the first network call. Staging a 200 MB reel to S3 and waiting a minute for ingestion, only to be refused on a caption length, is a bug.

• Receipts are per (platform, content). Editing the caption makes it a new post and it goes through; re-running the identical command does not.

• Report partial failure as partial. An X thread that publishes 3 of 6 posts is not a failed publish and must not be reported as one — the receipt records what actually went out.

• Verify endpoints before first use. All seven APIs move. Versions are env-overridable (INSTAGRAM_API_VERSION, LINKEDIN_API_VERSION, THREADS_API_VERSION, X_UPLOAD_URL) precisely so a vendor change does not need a code change. references/platform-limits.md carries the last-verified date.

• The browser path is a fallback, not a peer. It is best-effort, it is a grey area under some platforms' terms, and it requires explicit consent every single time. Never drive it silently.

• No engagement automation. This skill posts what the user wrote. It does not follow, like, comment, DM, or run multiple accounts in concert.

INVOCATION HINTS

Triggers:

• "опубликуй / выложи / запость это", "залей в инсту", "выложи карусель в треды", "кинь рил в тикток"
• "post this to X", "publish the carousel", "upload the reel to YouTube", "send it to the channel"
• Immediately after carousel-builder or reel-builder finishes, when the user says "и выложи" / "and post it"

Defaults: dry-run, no --draft, caption from captions.md. If the user names no platform, ask once — do not guess a destination.

If nothing is connected yet, route to references/oauth-setup.md and suggest starting with Telegram (no OAuth) or Threads (simplest OAuth), not Instagram.

Upstream: carousel-builder, reel-builder, viral-text, quote-card-maker, meme-card-maker — anything that lands assets in ./generated/.