Schedulala

Audience: the agent. A user has pointed you at this file ("read https://schedulala.com/SKILL.md and follow it"). Install and connect Schedulala, then use it to schedule and publish their social posts. Sections 1–3 get you connected; sections 4–9 are how to work once you are.

Schedulala schedules and publishes social media posts across 12 platforms, with analytics, engagement (comments/replies), and social listening. Full REST API, CLI, and connector docs: https://schedulala.com/developers/docs — link users there; do not restate the REST reference.

1. Get access

The fastest path on any agent with a shell is the CLI (zero MCP tool-schema cost). Node >= 18, no other dependencies:

npm i -g schedulala        # or prefix every command with: npx -y schedulala

The user gave you an API key (sk_live_ or sk_test_):

export SCHEDULALA_API_KEY=sk_live_...     # or: schedulala config set api_key sk_live_...
schedulala whoami --json                  # exit 0 = ready · exit 2 = missing/rejected key

Key resolution order: --api-key flag > SCHEDULALA_API_KEY env > ~/.schedulala/config.json. Keys are minted in the dashboard at https://schedulala.com/dashboard?view=developer (live or test). Pointing at a non-production server (self-hosted, staging): --base-url <url> or SCHEDULALA_API_URL env.

No key yet — run the device flow. Be honest with the user about the one human step it needs:

schedulala init --email their@email.com --json

This prints a code (like SCHD-XXXX-XXXX) and the URL https://schedulala.com/developers/verify. The USER must open that URL in a browser, sign in via magic link with the SAME email, and enter the code — you cannot complete this step for them. The CLI then polls (up to 15 minutes; codes expire in 15) and saves the key to ~/.schedulala/config.json. With --json, init emits TWO JSON documents on stdout (first the code + URL, later the completion) — parse them as a stream, never JSON.parse the whole output. Never ask the user to paste credentials into the chat.

What a fresh key gets (free taster): 2 connected accounts, 4 posts total — lifetime, never resets, 30 requests/min, no posting to X/Twitter (returns 402; connecting X still works). Existing Schedulala subscribers get their paid plan on the same key automatically. Upgrade link: schedulala upgrade --json. Test keys (sk_test_) simulate posts — nothing publishes, no quota burn — ideal for a dry run. Note: sandbox posts resolve IMMEDIATELY as posted with a sandbox id even when scheduled for later; that's the simulation completing, not your schedule being ignored.

2. Identify which agent you are

Token-sensitive runtimes: the full MCP surface is ~16k tokens of tool schemas, and the 9 show_* widget tools plus the attach_media widget path only render on claude.ai/ChatGPT. Filter client-side (OpenClaw toolFilter.include) to the core set — or use the CLI, which costs nothing.

3. Connect social accounts

Check what's connected first: schedulala accounts --json (MCP: list_accounts). To connect more:

• Scriptable, no browser: schedulala connect bluesky --identifier handle.bsky.social --app-password xxxx-xxxx-xxxx-xxxx and schedulala connect telegram --bot-token 123:ABC.
• OAuth platforms (twitter, instagram, facebook, linkedin, tiktok, threads, youtube, pinterest, google-business): schedulala connect <platform> --json returns an authorization URL — hand it to the user to open and approve; never try to complete OAuth yourself.
• Mastodon: dashboard-only (per-instance OAuth) — send the user to https://schedulala.com/dashboard?connect=1.

Verify afterwards with schedulala accounts --json, then you're ready to post (section 5).

4. Auth and safety rails

• Check auth before anything else: list_accounts (MCP) or schedulala whoami (CLI). No accounts / auth error → section 1 / section 3.
• OAuth, credentials, and payment happen ONLY on schedulala.com in the user's browser — never in the chat. Never ask for passwords, app passwords, bot tokens, or card details. connect_account and get_upgrade_link return links; that is the entire flow.
• Always confirm with the user before anything publishes (publishNow, a near-term scheduledFor, reply_to_comment, hide_comment). Prefer drafts and previews first.
• API keys: sk_live_ = real posting; sk_test_ = sandbox (simulated, no real posting, no quota burn). CLI: --sandbox requires an sk_test_ key.

5. Core publishing workflow

1. Discover — list_accounts: platforms, usernames, accountIds, health. isConnected: false + needsReconnect: true = dead token → tell the user to reconnect in the dashboard; do not post to it. When MORE THAN ONE account is connected on a platform, accountId is REQUIRED in the platforms array (the API returns a 400 listing the choices rather than silently picking). CLI: --platforms "bluesky=did:plc:xyz,twitter" — platform=accountId with the id from schedulala accounts --json. (REST equivalent: POST /api/v1/posts with platforms: [{platform, accountId}] — full reference at https://schedulala.com/developers/docs, don't guess other fields.)
2. Draft — write the content; per-platform text via the object form { platform, accountId, content } or platformSettings.<platform>.content.
3. Validate / preview — validate_post checks limits, media counts, and required fields per platform WITHOUT posting. On the hosted connector, prefer show_post_preview: a read-only, faithful per-platform mockup (threads, image grids, carousels, PDF docs, 9:16 reels, polls, live char counts). Pass it the exact draft you'd give create_post. (The CLI has no validate command — validate via MCP/REST, or create with --draft first and inspect.)
4. Confirm — show the user what will go out, where, and when.
5. Create — create_post with exactly ONE scheduling mode: scheduledFor (ISO 8601, 5 minutes to 90 days out) / publishNow: true / queue: true (auto-schedule next open slot) / status: "draft". CLI: schedulala post "<content>" --platforms <list> with --schedule <ISO8601> / --now / --queue / --draft (NO flag = draft) — full flags in references/cli.md.
6. Verify — get_post for per-platform status, URLs, and errors; retry_post re-runs failed platforms (optionally a subset, optional delay seconds). CLI: schedulala post:status <id> --watch.

Editing: update_post works on drafts and scheduled posts only (send only changed fields; mediaItems replaces ALL media). cancel_post deletes drafts and cancels scheduled posts; already-processing/posted posts can't be cancelled.

Bulk: bulk_create_posts takes up to 25 create_post-shaped entries. Threads (chains): create_thread, 2–25 entries, on twitter / threads / bluesky / mastodon only, max 1 media item per entry; the whole chain publishes from ONE postId — get_post that id for status.

First comment (auto-plug): pass firstComment (plain string) or firstComments ([{target, text}]) on create_post / bulk_create_posts / create_thread to auto-post a comment ~15 seconds after publishing — the "link in the first comment" workflow. Supported on linkedin, facebook, instagram, youtube, twitter, threads, bluesky, mastodon (NOT tiktok / pinterest / telegram / google-business — no comment API). Text and links only. Comment limits differ from post limits — see references/platforms.md. A failed comment never fails the post; get_post reports delivery per platform.

Quota: threads and bulk consume post quota PER ENTRY (a 5-entry thread = 5 posts). Validation and account-resolution failures don't burn quota. Check get_usage (or schedulala usage) when planning volume; on quota_exceeded or the sandbox plan, hand the user the link from get_upgrade_link.

6. Media decision tree

Full protocol and limits: references/media.md

• User's file (on device or attached in chat) → attach_media: on widget-capable clients (claude.ai) it opens an in-chat drop zone; on clients without widgets it still works — its result carries a programmatic presign→PUT→confirm upload API (see references/media.md). Full quality, images ≤20MB, videos ≤500MB, PDFs ≤100MB. NEVER read and base64-encode a user file yourself — chat channels truncate large payloads and destroy quality.
• Public http(s) URL → upload_media (mode url): re-hosts on the CDN (images ≤20MB jpeg/png/webp, gif ≤15MB; videos ≤50MB mp4/mov/webm). Videos LARGER than 50MB: skip upload_media and pass the public URL directly in mediaItems — it is fetched at publish time. Re-hosted media expires after 30 days if unused.
• Image YOU generated in this conversation (chart, card) → begin_generated_media_upload → append_generated_media_chunk (sequential seq from 0, ≤8000 base64 chars per chunk; 409 → resume from expectedSeq; 429 → wait, resend same seq) → finish_generated_media_upload. Images only, ≤3MB decoded, session expires in 30 minutes.
• Already uploaded ("my latest upload", "that reel from earlier") → list_media, then reference the returned url.
• CLI: --media <url[,url…]> (type inferred from extension).

Multiple mediaItems = carousel (instagram, facebook, threads, tiktok, pinterest; mastodon up to 4 images). PDF (type: "pdf") = LinkedIn document posts only.

7. Platform rules (the gotchas that reject posts)

Full per-platform schemas: references/platforms.md

Character limits: twitter 280 · bluesky 300 · threads 500 · mastodon 500 · google-business 1500 · instagram 2200 · tiktok 2200 · linkedin 3000 · telegram 4096 · facebook 63206. (YouTube and Pinterest count title/description via platformSettings instead.)

8. Analytics, engagement, listening

On the hosted connector prefer the widgets: show_analytics (one account's growth + top posts), show_leaderboard (ranked posts, sort engagement|likes|impressions|date), show_comparison (side-by-side 2–5 entities — YOU supply the metrics from the analytics tools), show_calendar (upcoming + recent posts), show_inbox (comments with Reply/Hide actions), show_feeds (keyword listening feed), show_connections (accounts + plan).

Data tools (all interfaces):

• get_post_analytics — per-post metrics + engagement rate. Platforms: instagram, twitter, bluesky, threads, facebook, linkedin, pinterest, tiktok, youtube, mastodon (NOT telegram / google-business). Omit accountId to merge top posts across all accounts (merged mode doesn't paginate).
• get_follower_growth — current count + 7d/30d growth + daily history. tiktok reports no follower counts; merged totals exclude YouTube.
• get_best_posting_times — top hours/days from engagement history.
• get_video_transcript / get_video_retention — YouTube only; retention has a 7-day cache and ~48h lag. Great for repurposing videos into posts.
• Engagement — comments + replies on facebook, instagram, youtube, linkedin, threads, bluesky, google-business; hide on facebook, instagram, youtube, threads; mentions on instagram, threads, bluesky. Reply targets differ per platform: instagram/youtube take commentId; facebook/linkedin take commentId or postId; threads takes either; bluesky requires all four replyRefs (parentUri, parentCid, rootUri, rootCid) from a comment's platformData.replyRefs. google-business "comments" are customer reviews — one owner reply, ≤4096 chars, replying again overwrites. reply_to_comment and hide_comment PUBLISH changes — confirm first.
• Listening — search_feeds searches public posts on bluesky + threads (threads has a Meta budget of 500 searches per rolling 7 days per account, shared with the dashboard — prefer bluesky for exploration); lookup_profile (bluesky only); list_feed_keywords / update_feed_keywords (max 20 per platform).

9. Errors → what to do

Worked examples (create, tiktok, multi-account, thread, bulk): examples/ in the installed skill, or https://github.com/schedulala/schedulala-agent/tree/main/examples when you fetched this file over HTTP.