Socialinsider

View saved version →

---
name: campaign-recap
description: Recap a social media campaign or content pillar using Socialinsider data — volume published, engagement earned, best posts, and how it compares to non-campaign output. Use when the user asks how a campaign, launch, hashtag or content pillar performed, wants a campaign wrap-up or post-mortem, or asks whether a content theme is working.
---

# Campaign recap

Report what a campaign or content pillar actually delivered.

## Rules

1. **Call `socialinsider_get_date_context` first, always.** Campaign questions are the most
   year-sensitive of all — never assume the current year.
2. **Resolve the campaign name, never guess it.** Call `socialinsider_resolve_campaign` with
   whatever the user said. It fuzzy-matches against the project's real tags and returns the best
   match plus alternatives. If the best match is not an obvious hit, show the user the candidates
   and ask which one they meant.
3. **Count server-side.** Use `socialinsider_count_posts` with `campaign_query` for volume. Never
   paginate `socialinsider_get_posts` per profile and add up the pages yourself.
4. **Read-only.**

## Steps

1. `socialinsider_get_date_context`.
2. `socialinsider_list_profiles` (and `socialinsider_list_campaigns` if the user has no name in
   mind and wants to see what exists).
3. `socialinsider_resolve_campaign` — turn the user's phrasing into the exact tag.
4. `socialinsider_count_posts` with the resolved `campaign_query`, the project, and the window.
   This returns the grand total, a per-profile breakdown, and any profiles that errored. Report
   errored profiles rather than quietly under-counting.
5. `socialinsider_get_top_posts` with `campaign_name` set, per profile that carried the campaign,
   `limit: 5`.
6. For the comparison baseline, run `socialinsider_get_posts_summary` twice per profile: once with
   `campaign_name` set and once without. The second is all output for the window, so campaign
   performance versus the profile's own everyday average is the difference between them.

## Output

- **Verdict** — one sentence: did the campaign beat the profile's normal output, and by how much.
- **Volume** — posts published, split by platform and profile. Note any profile that published
  nothing under the tag when it was expected to.
- **Performance versus baseline** — campaign average engagement per post against the window's
  overall average, as a percentage. This is the number that answers "did it work".
- **Top posts** — the best five, with the metric and the reason each landed.
- **What to repeat, what to drop** — concrete, tied to the numbers.

When the campaign window is short or the post count is small (under roughly ten posts), say plainly
that the sample is too small for a confident read and give the numbers without the verdict. Do not
manufacture a trend out of three posts.

View saved version →

---
name: competitor-benchmark
description: Benchmark a brand against its competitors on social media using Socialinsider data — share of engagement, posting cadence, engagement rate and format mix side by side. Use when the user asks how they compare to competitors, wants a competitive or industry benchmark, share-of-voice analysis, or asks who is winning on a platform.
---

# Competitor benchmark

Compare several profiles over the same window and say who is actually winning and why.

## Rules

1. **Call `socialinsider_get_date_context` first, always** — then reuse the exact same `date`
   object for every profile. A benchmark across mismatched windows is worthless.
2. **Same platform, same window.** Never compare an Instagram profile against a LinkedIn one in
   the same table. Group by platform and produce one table per platform.
3. **Never invent ids.** Resolve every profile through `socialinsider_list_profiles`. If the user
   names a competitor that isn't tracked, say so and stop — do not silently drop it and do not add
   it (adding profiles counts against their subscription; see the note at the end).
4. **Read-only** unless the user explicitly asks to start tracking someone.

## Steps

1. `socialinsider_get_date_context`.
2. Resolve the competitive set:
   - `socialinsider_list_profiles` for the project.
   - `socialinsider_list_brands` when the project is brand-grouped — brands are the intended unit
     of benchmarking, so compare brand against brand rather than a single profile against a brand.
3. Confirm the set with the user when it is ambiguous: list who you are about to compare, on which
   platform, over which window. One line, then proceed.
4. For every profile in the set, over the identical window:
   - `socialinsider_get_posts_summary` — post count, total engagement, average per post,
     engagement rate, post-type breakdown.
   - `socialinsider_get_profile_data` — follower count and growth over the window.
5. Optional, when the user wants to know *what* the competition is doing well:
   `socialinsider_get_top_posts` with `limit: 3` for the leaders only.

## Output

One table per platform, sorted by engagement rate (not by raw engagement — raw totals just
rank by follower count):

| Profile | Followers | Follower Δ | Posts | Total eng. | Eng./post | Eng. rate |

Then:

- **Ranking and gap** — where the user's brand sits, and the size of the gap to the leader in
  percentage terms.
- **Cadence versus efficiency** — separate the brands that win by posting more from the brands
  that win per post. These need opposite responses.
- **Format edge** — which post types the leaders use that the user's brand under-uses.
- **Two moves** — the highest-leverage changes, each grounded in a number from the table.

Always normalise by follower count before declaring a winner, and say so. If a competitor's
engagement rate is inflated by a single outlier post, name that post rather than the average.

If the user wants an untracked competitor added, state the exact profile URL, run
`socialinsider_profile_search` to verify it, present what will be created, and wait for explicit
confirmation before calling `socialinsider_add_profile` — it consumes their profile limit.

View saved version →

---
name: profile-audit
description: Audit how one social media profile performed over a period using Socialinsider data — followers and engagement trend, aggregate post metrics, post-type mix, and the posts that actually worked. Use when the user asks how a profile, account, brand or channel is doing, wants a performance review, monthly report, engagement-rate check, or asks why reach or engagement moved.
---

# Profile audit

Produce a performance audit for one tracked profile from Socialinsider data.

## Rules

1. **Call `socialinsider_get_date_context` first, always.** Your training data may be from an
   earlier year. Never guess or hardcode a year. Use the returned `date_ranges` values verbatim
   as the `date` argument (`{start, end, timezone}`, millisecond timestamps).
2. **Never invent ids.** `id` and `profile_type` must come from `socialinsider_list_profiles`.
   `projectname` is the project the profile lives in.
3. **Aggregate, don't paginate.** Use `socialinsider_get_posts_summary` for totals and averages
   and `socialinsider_get_top_posts` for rankings. Only fall back to `socialinsider_get_posts`
   when the user needs individual post records.
4. **Read-only.** This skill never calls a write tool (`create_*`, `add_*`, `remove_*`).

## Steps

1. `socialinsider_get_date_context` — get today's date and ready-made ranges.
2. `socialinsider_list_profiles` — resolve the profile the user named to its `id`, `profile_type`
   and `projectname`. If several profiles match, ask which one before continuing.
3. Pick the window. Default to `last_30_days` when the user didn't say. If the user asks about a
   trend or change, also fetch the immediately preceding window of equal length so you can compare.
4. Gather, in parallel:
   - `socialinsider_get_profile_data` — daily followers, engagement rate, reach, impressions.
   - `socialinsider_get_posts_summary` — true post count, totals, averages, breakdown by post type.
   - `socialinsider_get_top_posts` — `metric: "engagement"`, `limit: 5`.
5. If the profile is `instagram_profile_own`, add `socialinsider_get_instagram_stories` for the
   same window. Skip it silently for any other profile type — it is unavailable there.

## Output

A short written audit, in this order:

- **Headline** — one sentence: direction and size of the move ("engagement rate down 18% MoM on
  flat follower growth").
- **Numbers** — a compact table: followers (start, end, delta), posts published, total engagement,
  average engagement per post, engagement rate. Add the prior-period column when you fetched one.
- **What worked** — the top posts, with the metric that made them top and what they have in common
  (format, topic, length, posting time). Name the pattern, not just the posts.
- **Post-type mix** — which formats the profile leans on versus which formats earn engagement.
  Call out a mismatch when it exists.
- **Recommendations** — at most three, each tied to a number you just showed.

State the date range and timezone you used. Flag any metric the API returned as empty rather than
reporting it as zero — for non-owned profiles, reach and impressions are often simply unavailable.