Installable skills for Claude and other agents, published by Scout7.
Download a .skill file and open it in Claude (Save skill), or copy the
unzipped folder into ~/.claude/skills/. Agents can discover these skills
automatically via index.json
(Agent Skills Discovery RFC).
scout7-run-organic-loop
Runs a brand's organic growth loop on Scout7 — plans, produces, and publishes image posts, videos, blogs, and LinkedIn carousels, then reports the impact. One approval up front; the rest runs unattended — or in manual mode, each piece pauses for your OK before credits are spent.
What it can do
Audit the connected Scout7 workspace — topic coverage, calendar, growth state — and top up topics when coverage is thin.
Interrogate you — researching the brand and market on the web — to pin the loop's focus, your philosophy, and your target, then record it in the brand's canonical facts so the topics chosen and every angle drafted all come from it.
Draft a 1-day or 1-week content plan (topics, angles, platforms, timing, and the real credit cost) and refine it with you until you approve.
Lock every decision into a plan file on disk, so the plan — not a human — answers every question the run hits later.
End planning with your choice of run mode: fully autonomous, or manual — where each blog outline, post concept, and video script comes back to you for approval before anything is generated, and the finished pieces return for a final OK before they're scheduled.
Generate each piece with Scout7, wait out the async renders, and end every run by showing you the finished content itself — copy, images, blog, video — then schedule or publish to your connected accounts exactly as approved; anything you mark draft-only stays a draft.
Register its own daily Executor and Reporter jobs with a durable scheduler, so creation and reporting fire without anyone watching.
Answer blog author-interview questions from the brand's recorded facts/ and stance/ files — it never invents the founder's voice.
Write an end-of-day impact-and-learnings report from live Scout7 metrics, closing the loop that feeds the next day's plan.
What it needs
A connected Scout7 workspace (MCP).
A brand knowledge folder with facts/canonical.md + stance/*.md — bootstrapped automatically from Scout7 brand context on first run.
A durable scheduler for the unattended stages (e.g. scheduled tasks in Claude/Cowork).
Sets up the always-on engagement engine for a brand: verified social logins, a short voice interview, and one recurring routine per platform that keeps commenting, replying, liking and following in that brand's voice — proven to be firing before the setup is called done.
What it can do
Get you signed in first: you log in yourself (it never touches credentials), and it reads the handle off the screen to confirm the account really is this brand's — the check that prevents ever posting as the wrong person.
Ask nine questions, once. Who is writing, what may be claimed as true, what is off-limits, how long a comment runs, what to do when someone is wrong, how often to name the product, and the words you would never use. Anything your Scout7 brand context already knows is confirmed rather than asked.
Turn those answers into the three files that decide whether an unsupervised comment reads as a person: the canonical facts it may claim, the voice it writes in, and the phrases that give a bot away. It drafts them, shows them back, and takes your corrections.
Generate a standalone engine folder with measured defaults for daily targets, ramp-up, quiet hours and per-platform pacing — across 11 platforms, each with its own playbook of hard-won mechanics.
Register one recurring routine per platform, then prove each one actually fires before handing over. A health check reports pass/fail per platform and names the single next action; setup is not finished until every routine has logged a run.
Keep running without you. Steering afterwards is conversational — "that comment was too salesy", "never engage that account", "pause X for the weekend" — never file editing.
What it needs
Claude for Chrome, so the scheduled runs reach your logged-in accounts.
Python 3.9+ (no packages to install) and a scheduler.
You at the keyboard for the logins and the nine questions — about ten minutes.
The engine stays paused until you say go, after a supervised first cycle.
Teaches your agent what Scout7 can actually do — from the live tool list, not from a summary — tells you what you can ask for next in terms of your own brand, and writes it into the agent's memory so it never has to relearn it.
Why it exists
Scout7 is around 96 tools. An agent that was never briefed on them treats it as a lookup table: it does exactly what you asked and never mentions that the article it just wrote could become a LinkedIn carousel, that a preview would have shown the cost before spending, or that you can check whether ChatGPT cites your brand at all. This runs once at setup and closes that gap permanently.
What it can do
Read the tools your connection actually registered, so the briefing is true for your grant — it never promises a capability you don't have, and never hides one you do.
Read your workspace first, so what you hear is about your brand and your current state: no topics yet gets a different briefing than already-publishing.
Tell you, in plain language, the handful of things worth asking for next — and what's connected versus still missing, like publish identities or your video face and voice.
Write two files with your permission: a full capability reference on disk, and a short pointer in the file your agent loads every session. Split that way on purpose, so the reference doesn't eat your context budget in unrelated sessions.
Stay honest as Scout7 grows: a coverage audit reports any live tool the map doesn't yet describe, so the briefing can't silently fall behind the product.
What it needs
A Scout7 MCP connection.
Python 3 (no packages to install).
Somewhere to persist — a memory directory or a CLAUDE.md. Without one it still briefs you, just for the session.
Re-running is always safe: it replaces its own block rather than appending, so nothing duplicates and nothing else in your files is touched.