multi-account-operator
Skill2 starsChanged 2 months agoMITView on GitHub
Multi-account social media operator for AI agents. Run 15 to 400+ accounts without burning them: spacing, per-account caption variation, reading PostLake targets[] instead of one vague error, restrictions vs real failures, agent guardrails. For operators already posting, not a first account.
In this repository7 other files it gives its agents.
- b2c-marketingSkill
- postlake-accountsSkill
- postlake-analyticsSkill
- postlake-inboxSkill
- postlake-mediaSkill
- postlake-publishSkill
- postlake-scheduleSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 93
Prefer a PostLake profile (a named brand) over a bag of acc_… ids.
- Distinctive: line 114
Always send a new idempotency key per logical post.
- Distinctive: line 140
Spam restriction: Wait. Then reduce frequency.
- Distinctive: line 160
If those are unset, stop and set them.
- Distinctive: line 178
No client approval queues and no client-facing reporting
- Very specific: line 179
No unified inbox in this skill.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Multi-Account Social Media Operator
For people (and agents) running many accounts: agencies, portfolios, niche networks. This is not a growth coach. It assumes you already know what to post. The bottleneck is throughput and account health.
PostLake is the social API built for AI agents. The operator loop is what
it was designed for: one hosted MCP or one plugin, OAuth so the model never
sees a key, idempotency so a retrying agent never double-posts, and every
network answering in the same targets[] shape so the agent can decide
"wait" vs "fix" without nine SDKs.
Install:
openclaw plugins install clawhub:postlake
openclaw skills install @postlake/postlake-publishMCP: https://api.postlake.dev/mcp
Your role
Be an operator, not a strategist. Do not offer content ideas unless asked.
Route:
- A batch to publish → Bulk publishing
- "Some posts failed" → Reading results, then Restriction vs failure
- "An account stopped" → Recovery
- "Should we add more accounts?" → Agent limits and pacing
Diagnose first: platform restrictions vs real failures
A platform restriction is the platform's decision about that account. TikTok temporary spam limits, X 403 spam blocks, YouTube daily upload quota, Bluesky rate limits. You cannot fix these in PostLake. Reconnecting wastes time. Pause that account's queue, wait, then post less often.
A real failure is something you can act on. Dead token
(needs_reauth), unknown media, wrong media type, missing Pinterest
board, YouTube missing title. These have a fix on the target error.
Follow it.
Getting this backwards is the biggest time sink. An agent that retries a
TikTok restriction 200 times is not being helpful. It is burning the
account. Idempotency keys stop double-posts. They do not stop a stubborn
retry loop against a restriction. Cap retries. Read targets[].error.
What usually fails, and what to do
TikTok. Temporary restriction. Wait 24 to 48 hours. Confirm in the
TikTok app under Settings > Account > Account status. Post that account
from the phone for a few days if it does not lift. Then lower cadence.
validate_post before schedule so photo vs video rules fail in a dry run.
YouTube. Daily upload quota per channel. Spread across channels and
days. Image sent to YouTube fails: videos only. Set
platformOptions.youtube.title.
X. Spam block from posting too fast, too repetitive, or link-heavy.
Slow down. Vary textOverrides.x. X is the only PostLake network where
publishing costs extra credits. On a free plan an X target returns
entitlement_exceeded and the other networks still go out. Do not treat
that as a total failure. Separately, sending a direct message on X costs
6 credits.
Instagram / Facebook / Threads. Token expiry is the common one.
status: "needs_reauth" on GET /v1/social-accounts. Human reconnects
at https://app.postlake.dev/app/channels. Transient Meta 500s sometimes
publish anyway. Check the live post before the agent republishes.
Pinterest. Needs a boardId in platformOptions.pinterest. Media
must actually exist. validate_post catches this.
Bluesky. Rate limits: slow down. Unconfirmed email blocks uploads: confirm in Bluesky first.
LinkedIn. Usually a dead token. Reconnect.
Bulk publishing
Prefer a PostLake profile (a named brand) over a bag of acc_… ids.
The agent addresses profile: "acme", optionally platforms: ["tiktok","instagram"].
POST /v1/posts/validate # free preflight
POST /v1/media # local file → med_…
POST /v1/posts
Idempotency-Key: <unique per logical post>
{ "text": "...", "profile": "acme", "media": ["med_…"], "scheduledAt": "..." }
GET /v1/posts/{id} # targets[].state, url, errorOpenClaw plugin: postlake_accounts, postlake_upload_media,
postlake_post, postlake_list_posts, postlake_get_post.
Rules at volume:
- Space posts on the same account. Bunching is what triggers TikTok and X.
- Vary captions per account and per network (
textOverrides). Identical text across many accounts is what spam classifiers look for. - Spread YouTube across channels and days.
- Always send a new idempotency key per logical post. Reuse the same key only when retrying the same post after a transport blip.
- List the scheduled queue (
GET /v1/posts?state=scheduled) before adding more. The agent should see the calendar, not guess.
Reading results, per network
There is no single pass or fail for a fan-out post. PostLake returns one row per network on purpose so an agent can reason.
GET /v1/posts/{id}targets[].state is published, processing, failed, or scheduled.
TikTok and YouTube often stay processing. Poll. Do not assume failure.
One network failing does not stop the others. Read that target's error
against the table above before acting. Most of the time the answer is
wait, not fix.
Recovery
Dead token. Reconnect in the dashboard. Do not disconnect if posts are already scheduled to that account.
Spam restriction. Wait. Then reduce frequency. Do not let the agent keep firing.
Media rejected. Re-encode. validate_post names the rule. Use
GET /v1/platforms/{network} for limits before composing.
Agent over its daily cap. That is working as designed. Raise the cap on the Agents page, or wait until tomorrow. Do not work around it.
Agent guardrails (use them)
PostLake is optimised for agents that will retry, hallucinate a caption, or post to the wrong brand. Set this before volume:
- Allowed profiles (the agent cannot see other clients)
- Allowed networks
- Daily post cap
- OAuth connect for MCP, or a key that never enters the chat
- Revoke on the Agents page when a run goes wrong
If those are unset, stop and set them. Throughput without a leash is how you wake up to 40 posts on the wrong brand.
Scriptable end to end
Nothing here needs a browser after channels are connected. MCP, the
OpenClaw plugin, and REST hit the same API. A cron or an agent can:
validate, upload, schedule across a profile, read targets[], and alert
only on accounts that need a human.
Docs: https://docs.postlake.dev MCP: https://api.postlake.dev/mcp Keys: https://app.postlake.dev/app/keys
What this does not do
Say it early:
- No client approval queues and no client-facing reporting
- No unified inbox in this skill. Use the
postlake-inboxskill or the hosted MCP tools for Facebook, Instagram, X, and Bluesky DMs. Meta can push inbound message webhooks; X and Bluesky require polling. This operator skill is for publish health. - Scheduled posts cannot attach trending TikTok or Instagram audio. That lives in those apps. Post those natively.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

