Review performance

Review Performance

Find out what is actually working on your socials and what to do more of. This skill reads your real engagement numbers through the Marky API, looks for patterns, and gives you specific, data-backed recommendations.

Read the marky-api skill first for the base URL, your mk_live_ key, and the full endpoint list. Everything below uses https://api.mymarky.ai/api and the header Authorization: Bearer mk_live_YOUR_KEY.

Marky wants your feedback. If anything breaks or is confusing while you run this skill — and again once you finish — send Marky a quick note with the submit_feedback MCP tool. See the "Marky wants your feedback" section in the marky-api skill for when and how.

Everything here uses only read endpoints, so it is safe to run anytime.

Cut by metadata tags

Posts created by agents carry metadata (media_type / format / hook / topic — see "Tag every post" in the marky-api skill). When enough tagged posts have stats, group engagement by each key and report which values win — "diagrams out-engage photos 2:1", "question hooks beat bold claims". This is the highest-signal section of the review; lead with it when tags exist, and note when the sample is too small to trust (<5 posts per bucket is directional only).

Step 1 — find your business and connected accounts

Two MCP tools: list_businesses (copy the id you want as business_id), then list_connected_social_accounts on that business (copy each integration id, and note its platform and status).

If no accounts are connected, stop and tell the user to connect a social account in the dashboard first. Stats only exist for connected platforms.

Step 2 — pull the numbers

Pull three things: account-level stats, the posts published on each platform (with their engagement), and per-post detail when you want to go deeper.

• get_integration_stats(business_id, integration_id) — account-level audience stats (followers, growth) for one connected account.
• list_integration_posts(business_id, integration_id) — posts published on that platform, each with its engagement numbers. This includes posts published OUTSIDE Marky, so it is the true account picture; get_external_post_stats drills into one of those external posts.
• get_post_analytics(business_id, post_id) — engagement for one specific Marky post (across the platforms it went to).

Repeat the account/posts calls for each connected integration so you can compare platforms.

To tie performance back to what you posted about, also list your posts and topics:

list_posts(business_id) for your post history (filter by status if you like), and list_topics(business_id) so you can group results by topic.

If the business has a Google Business profile connected, you can also pull customer reviews to round out the picture (reviewer, star rating, text, and any reply):

list_google_reviews(business_id, order_by="rating desc") — newest first by default; order_by also takes rating.

Step 3 — find the patterns

Compare across a few dimensions and write down what stands out:

• Top posts. Which 3-5 posts got the most engagement? What do they have in common?
• By topic. Group posts by their topic_id and compare average engagement. Which topics pull their weight and which are dead weight?
• By format. Compare image posts vs video vs text. One format usually wins.
• By platform. Average engagement per platform. Some platforms reward different content.
• Follower growth. Is the audience growing, flat, or shrinking on each platform?

Step 4 — recommend and (optionally) act

Present a four-tier action plan. Lead with the result, not the metric, and cite the specific posts behind every recommendation — a rec with no post behind it is a guess:

• Quick wins — do this week, low effort ("Your how-to posts get ~3x the engagement of your promo posts — post two more how-tos").
• Strategic shifts — change the mix ("Video outperforms image on TikTok but image wins on LinkedIn. Match format to platform").
• Experiments — worth testing, not yet proven ("Topic 'client wins' is your best performer on 2 posts — small sample; run 3 more before concluding").
• Stop doing — the negative space most reviews skip ("Hashtag-wall captions are your bottom quartile on LinkedIn — drop them").

Benchmark against the business's OWN averages, not platform folklore — "good" is whatever beats their last 90 days. Flag every thin sample (<5 posts per bucket is directional only).

Then offer to act on it with the writing skills:

• Save the learnings. Offer to append the data-backed findings to the business's performance-learnings.md (see "Performance learnings" in the marky-api skill for the file's home and format). Creation skills read it before drafting, so a saved learning steers every future batch. Show the exact lines first; write only on a yes.
• Add or sharpen topics with the suggest-topics skill.
• Produce the next batch with the schedule-posts or plan-social-content skill, leaning into the winning topics and formats.

Notes and limits

• The API gives you per-post and per-account engagement plus your post and topic lists. Group and compare those yourself, the API does not return pre-aggregated leaderboards.
• Posts are linked to a topic_id, so you can group by topic. Categories now have their own endpoints too (GET /businesses/{id}/categories), so you can roll topics up by category if you want a higher-level cut.
• Stats are read-only here. Acting on the findings (new topics, new posts) happens through the other skills.