Performance review

Find out what worked on the user's accounts and turn it into next steps. This workflow only reads data, so any workspace role can run it.

Ground rules

• Never invent numbers. Every metric you quote must come from a tool result in this conversation. If a tool returns nothing, say that the data is missing and do not estimate it.
• The explicit account. Take socialAccountId from list_accounts and pass it on every call. With only a platform, Artima picks the first account on that platform. If the user has several accounts on one platform and did not say which, ask.
• Errors can look like results. analyze_post_performance and list_posts can return {"error": "..."} as a normal result. Check the payload before you analyse it.
• Engagement rate beats raw counts. A post shown to more people collects more likes. Rank by engagement_rate, and say that you did. Artima gives it as a fraction of impressions: 0.045 means 4.5%, (likes + comments + shares) / impressions.

Workflow

1. Pick the accounts and the scope

list_accounts, then settle which accounts and roughly how many recent posts to look at. "Last month" usually means 15 to 30 posts. When the user names no scope, use the latest 10 to 20 posts and say so.

2. Get the aggregate in one call per account

Call analyze_post_performance with socialAccountId and limit between 5 and 30. It returns:

• analyzed: how many posts had metrics;
• aggregate: totals and averages, including avgEngagementRate (a fraction);
• top and bottom: 3 posts each, sorted by raw interactions (likes + comments + shares), not by rate;
• all: one row per analysed post, with postId, likes, comments, shares, impressions, reach, engagement_rate, the first 140 characters of text, and publishedAt.

Re-rank all by engagement_rate yourself. A post can lead top on sheer reach while converting worse than smaller posts, and that is often the most useful thing to point out. This one call covers the whole set, so do not call get_post_analytics once per post to rebuild it. Keep get_post_analytics for a single post the user asks about.

Read the result honestly:

• analyzed: 0, or a note saying there are no metrics: the account's metrics have not synced yet, or nothing was published. Say exactly that, offer to list what was posted (step 4), and suggest checking again later. Do not produce rankings.
• analyzed lower than limit: either the account has fewer published posts, or posts without metrics were skipped silently. Say how many posts the analysis covers. If list_posts shows more published posts in the period, say how many were left out.
• {"error": ...}: report the error.

3. Compare in depth when it helps

compare_posts takes 2 to 6 postIds. Any other count is rejected, so split larger sets. When a post in the result carries an error entry, that post has no metrics. The call itself did not fail.

4. Add what the posts actually said

Each all row carries the first 140 characters of the caption, which is usually enough. When you need the full text of a post you discuss, call get_post_detail with its postId.

To browse what was posted in a period, call list_posts once per account. If the user named no period, ask for one. Pass either range (day, week, month or list, anchored by referenceDate) or startDate with endDate, never both. It reads Artima's calendar, drafts included, so keep only status: "Posted" for performance work. statusCounts covers the whole window, while posts is a sample (default 25, maximum 100).

5. Find the patterns

Look across the top and bottom posts for:

• format: post, carousel, reel or video, and whether media was attached;
• hook: the first line and the question or claim it opens with;
• topic and angle: product, behind the scenes, education, offer, social proof;
• length and CTA;
• timing: day and hour. publishedAt is UTC, so convert it to the user's timezone. If you do not know their timezone, ask or state the times in UTC.

With fewer than about 8 analysed posts, call the patterns tentative.

6. Report

Use this shape:

1. Headline: one or two sentences, such as "Behind-the-scenes posts got about 2× the engagement rate of product shots."
2. Table of the top and bottom posts: first line of the caption, publish date, likes, comments, shares, reach and engagement rate, using only the fields that exist.
3. 3 to 5 patterns, each with the evidence behind it.
4. 3 to 5 recommendations, each concrete: what to post, which format, which angle, and an example hook. Write them in the user's language.
5. Caveats: skipped posts, a small sample, missing metrics.

Then offer the next step without taking it, for example "Shall I draft next week's posts around these ideas?". Creating drafts belongs to the weekly-content-plan skill and needs the user's go-ahead.