--- name: divergence description: Refuses generic AI ideas. Generates non-obvious creative angles for content, ads, campaigns, hooks, videos, posts, and positioning by attacking its own first drafts and forcing reframes. Use when brainstorming, ideating, or asking for angles/concepts/ideas — especially when the first ideas feel obvious, safe, or like something any AI would say. Invoke with /divergence, "give me angles on X", "brainstorm X", "ideas for X", or "make this less generic". --- # Divergence — the anti-slop layer for *ideas* Most AI brainstorming is slop. Ask for ten ideas and you get the ten most statistically obvious ones — the same ten everybody else's AI gave them. That's not ideation, it's autocomplete. Divergence fixes that with one move: **generate, then attack your own output, then reframe what fails.** Never hand over the first draft. > `/human` cleans up *language*. Divergence attacks *ideas*. Different jobs. Use both. ## The rule that makes this work **Never show the user a list of formats or example angles to pick from.** Examples in a prompt become the ceiling, not the floor. The moment you offer "you could do a listicle, a before/after, or a testimonial," you've capped the thinking at what already exists. If you catch yourself reaching for "consider trying…" or pulling a known format out of memory — stop. That's the slop trap. --- ## Process ### Step 1 — Generate candidates with full latitude Generate 10 candidate angles on the topic. Use whatever context you have: the brand, the audience, the product, the data, what's actually true. **Zero example formats. No genre lists. No "here are some popular approaches."** If `brand.json` exists in this skill folder and is filled in, load it — real brand context makes the angles sharper. If it's empty, work from the topic and whatever the user told you. Don't block on it. ### Step 2 — Attack every candidate against the slop floor Run each candidate through this. Any hit gets tagged `slop`: - Is it a talking-head testimonial? → **slop** - Is it a before/after? → **slop** - Is it a listicle ("3 things I wish I knew")? → **slop** - Does it open with "POV:" / "Imagine if" / "Ever wondered" / "Picture this" / "Did you know"? → **slop** - **Could this exact angle belong to any competitor in the category?** → **slop** (this is the sharpest test — use it hardest) - Does it lean on a worn trope — "morning routine," "what's in my bag," "get ready with me," "day in the life"? → **slop** - Does the brand/product show up before the audience has been given anything of value? → **slop** - Is it aspirational lifestyle that quietly signals "you'll never have this"? → **slop** - Does it claim a result without any specificity? → **slop** - Would this idea have been equally obvious to anyone who thought about the topic for 15 seconds? → **slop** Be ruthless. Most first-draft candidates *are* slop. That's expected and it's the point. ### Step 3 — Force reframes on everything you tagged For each slop-tagged candidate, generate 3 reframes using these transforms. These are **rules to apply**, not examples to copy: - **Inversion** — flip the assumption. If the obvious version says "here's what to do," the inverted one says "here's why the way you've been doing it actually works, and when it doesn't." - **Contrarian** — argue the unpopular position. If everyone agrees on X, find the real case where X is wrong. - **Cross-domain metaphor** — connect the topic to a completely unrelated field. Architecture, weather, cooking, music, language, sports. The collision is where the idea lives. - **Specificity drilling** — instead of generalizing, drill into one ultra-specific case the audience will recognize instantly. The narrower it gets, the more people feel seen. - **Direct eye contact** — stop broadcasting. Write as if you're talking to one specific person you have in mind, about a thing only they would notice. Reframes don't have to be perfect. They have to break the pattern and give the user real material to work with. ### Step 4 — Score honestly and cut Score every candidate and reframe out of 10 on three axes: - **Slop resistance** — how unlike generic AI output is this, really? - **Fit** — does this actually work for this brand, audience, and product? - **Specificity** — is there a real, particular observation in here, or is it a shape with nothing inside it? **All three must be 6 or higher.** An angle that scores 9 on specificity and 4 on fit is dead — kill it. Rank by the *lowest* of the three scores, not the average. A great idea that fails one axis is a bad idea. Return the top 3. ### Step 5 — Name the emotion Every angle must evoke one of these, and you have to be able to name which: **recognition** ("that's me") · **curiosity** (an open loop) · **quiet validation** ("I wasn't crazy") · **mild surprise** (a small reversal) · **shared complaint** ("finally someone said it") · **tension** (something at stake) If you can't name the emotion, the angle isn't finished. Go back. --- ## Output format ```markdown # Divergence — {topic} ## Top 3 Angles ### 1. {short name} **The angle:** {one or two sentences} **Why it isn't generic:** {one sentence — be specific} **Emotion:** {recognition / curiosity / quiet validation / mild surprise / shared complaint / tension} **Scores:** slop resistance {n}/10 · fit {n}/10 · specificity {n}/10 ### 2. {...} ### 3. {...} ## Killed (the audit trail) - {candidate} — {why it was slop} - {candidate} — {why it was slop} ``` **Always show the kill list.** It proves the work happened, and it stops the user from re-suggesting the ideas you already rejected. --- ## Hard rules 1. **No format hints in generation.** Brand, audience, topic, truth. Nothing else. Never "here are some formats to consider." 2. **Self-attack before output.** Never return a candidate that hasn't been through the slop floor. 3. **All three scores ≥ 6.** No exceptions. Rank by the minimum score, not the average. 4. **Name the emotion or it isn't done.** 5. **Show the kill list.** 6. **Use the five transforms.** Don't invent new ones mid-run. ## What this skill doesn't do - It doesn't write the finished script, post, or ad — it produces the *angle*. Hand the angle to `/human` to write it. - It doesn't fix language. That's `/human`. - It doesn't hold a library of "good formats." No such library exists here, on purpose. ## Optional: brand context `brand.json` in this folder is optional. Fill it in and the angles get noticeably sharper, because the skill stops guessing at who you're talking to. Leave it empty and it still works from whatever you tell it in chat.