More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Outlier Post Finder: What Overperformed and Why

    1

    Find which of your own posts beat your account's normal numbers, measure the lift against an honest median baseline, diagnose why the winners worked, and bank their hooks.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Outlier Post Finder: What Overperformed and Why

    Outlier Post Finder: What Overperformed and Why

    Example session with this skill installed

    I run a home-cooking Instagram account. Here are my last 24 reels from the past 60 days (views): 3.1K, 4.2K, 4.8K, 5.1K, 5.4K, 5.6K, 5.9K, 6.1K, 6.3K, 6.6K, 6.9K, 7.2K, 7.5K, 8.1K, 8.8K, 9.4K, 10.2K, 11.5K, 12.8K, 14.9K, 18.6K, 27.4K, 41.8K, 96.5K. The 96.5K one opens with me dropping a whole tray of meatballs and saying "I wasted $40 of groceries so you don't have to." The 41.8K one is a fridge-cleanout pasta starting "Everything in this pan was about to be thrown out." The 27.4K is a knife-skills basics reel. Which posts actually matter, and what do I repeat?

    • Read your context and instructions
    • Compiled the outlier post finder

    Baseline first: 24 reels, one account, one format, 60 days — full confidence. Median views: 7,350 (not the 96.5K-flattered mean of ~13.9K). Lift, ranked: 96.5K = 13.1× (remarkable), 41.8K = 5.7× (remarkable), 27.4K = 3.7× (strong), 18.6K = 2.5× (strong). Everything at 14.9K and below is under 2× — normal range, whatever the raw totals suggest. Your biggest post is also your biggest outlier here, but the knife-skills reel matters more than its rank suggests, for a reason below.

    Step 5 read on the top two: both open mid-disaster with a loss stated in the first sentence — "$40 of groceries wasted," "about to be thrown out." Both are rescue narratives: something going wrong or going to waste, fixed on camera. Diagnosis scores: hook specificity 2/2 both, tension 2/2 both, repeatability 2/2 both. Two remarkable outliers sharing the same structure is a pattern, not a hunch:

    the loss-and-rescue opener is your account's engine.

    The knife-skills reel (3.7×) is the second signal: it's your only pure-technique post in the top four, and basics content is infinitely repeatable. It doesn't share the rescue pattern, so it goes in the ledger as a strong hunch: technique basics outperform your median by roughly 3–4×.

    Hook bank, verbatim: "I wasted $40 of groceries so you don't have to." / "Everything in this pan was about to be thrown out."

    Next experiments: (1) a "rescue" series — one ruined-or-wasted ingredient per week, loss stated in the first two seconds; success = median lift of the series above 2×. (2) Two more technique-basics reels (pan temperature, seasoning timing) to confirm or kill the hunch. (3) A rescue + basics hybrid: "You're throwing out the best part of the chicken" — tests whether the patterns stack.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    "What's our best post?" usually names the oldest viral hit or the biggest raw number — neither tells you what to repeat. This skill asks the right question: which posts did unusually well for this account, and what made them work?

    The method is numbers first, interpretation second. You scope tightly — one account, one platform, one format, one window — because reels and carousels, TikTok and LinkedIn, are different sports with different physics. The baseline is the median, not the mean, so one old hit can't make everything else look like failure. Every post gets a lift score (its metric divided by the baseline median), sorted into a decision table: remarkable at 5×, strong at 2–5×, mild below that. The biggest post is often not the biggest outlier, and this skill will say so every time.

    Then the winners get read properly: the exact first words quoted verbatim (never invented), the format, the topic, the tension that made it hard to scroll past, the timing — plus the comments mined as a second signal, where confusion is a content idea and buying intent is a product signal. A five-dimension diagnosis scorecard forces the read to be evidence, not vibes, and a pattern only earns its name with at least two posts behind it. The report you get has the ranked outlier table, a pattern ledger with evidence attached, a hook bank of real opening lines, and 3–5 next experiments, each one naming the pattern it tests. The same lens runs on a competitor's account — borrow their structures, never their persona.

    Edge cases are covered: mid-window pivots, boosted posts that teach you nothing, seasonal spikes, deleted posts quietly inflating your median. Confidence is labeled honestly: under 20 posts per baseline, everything is directional and the report says so.

    The skill analyzes data you supply or can access; it doesn't scrape platforms for you, and private or deleted posts won't appear in exports — the method flags that gap but can't fill it. Diagnosis quality depends on access to the actual posts and comments; metrics alone produce lift scores, not explanations. With fewer than 20 posts per baseline the output is explicitly directional, and no method fixes a nine-post sample. It also can't separate a genuinely better post from an algorithmic push the platform gave it for reasons outside the content — externally boosted posts are flagged and excluded from patterns rather than explained.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    Permissions

    No special permissions declared or detected

    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    No — and the skill will refuse to blend them. Baselines are per account × platform × format, so you run it once per platform and compare the patterns across reports. The patterns are the portable part; the numbers are not.

    You can run it, but the report will be labeled low confidence and every pattern will be a hunch. That's still useful for deciding what to test next — just don't retire a format on twelve posts of evidence.

    Yes — public metrics (views, likes, comments) are enough to build their baseline and lift table. You lose saves and shares, so the diagnosis leans on hooks, format, and comments, which are public anyway.

    New in Marketing & SEO