SuperAPI AI Directory

angle-mort

General & other

View repository

Updated 10-16 05:47

About this skill

blind spot — flip the obvious framing

A divergenceengine. You switch it on in the middle of a reflection to force new perspectives: it takes the obvious framing of a topic and flips it along several axes, until the angles you couldn't see appear — the blind spots.

Its law, non-negotiable: dare a lot, validate nothing. The skill produces bold angles, it never validates . Each angle comes out with a status label so you know at a glance what needs a check and what remains a thought experiment. Verification is another floor (a fact-check tool, or you).

It is generalist by construction : it knows nothing about the domain. A blind spot on "the electric car" and a blind spot on "should I accept this job" go through the same engine. That's intentional — as soon as you stick a domain lens or an audience guardrail on it, you bridle the divergence, and it's precisely the weird angle that often makes the good idea.

When to use it

Explicit triggers: "blind spot", "flip that", "inverse angle", "counterpoint", "what angle are we missing", "see what I'm missing", "blow up my blind spots".

Self-invocation (light): mobilize it on your own when the user keeps rehashing a single framing, asks the same question in three forms, or is about to execute / decide without having looked at the subject along another axis. Not on every message — only when the enclosure is real. Offer it ("shall I do a blind spot on that for you?") rather than imposing it.

The two inputs

The skill detects on its own how it is called:

  • Implicit context (default). Invoked without a specified topic, it takes what was just discussed and flips the framing of the moment. This is the main use: "give me a blind spot on what we're talking about".
  • Explicit topic. Invoked with a topic ("blind spot on remote work"), it starts from that topic cold, without depending on a history.

The engine: 5 core moves + 3 support moves

Each angle is born from a move, a type of flip. The details, sub-variants and examples are in references/moves.md — read it before a run if you're unsure about a move.

Core (the 5 basic ones, to sweep through first):

  1. Direction inversion — backbone. Take the natural direction, invert it: filter↔generator, result↔method, why↔how, restrict↔expand, or break the hidden premise (invert the unspoken assumption of the question, not its direction).
  2. Temporal shift (shift-left) — shift the timing: what if we handled this upstream rather than patching downstream?
  3. Nail test — is this really the right tool / the right frame, or are we over-applying something we like?
  4. Change of scale — zoom in to the micro (one case, one detail) or zoom out to the macro (the system, the decade); the subject changes nature.
  5. Inversion of viewpoint — look from the other side: the skeptic, the competitor, the beginner, the opponent, the one who loses.

Supporting moves (secondary — we don't include them by default, we draw them when the core is saturated or the subject calls for them):

  1. Fusion / combination — glue the subject to an unrelated field (subject × another domain / another object) and see what comes out of the collision.
  2. Subtraction / extreme constraint — remove the central element (“what if it didn't exist?”) or impose a brutal constraint (zero budget, 10 minutes, forbidden, only once).
  3. Analogic transfer — how another discipline (medicine, aviation, gaming, logistics) solves structurally this type of problem, then bring the mechanism back.

Out of scope, deliberately: check-the-check (turning the audit back on itself) and watch-your-thinking (extracting one's own thought patterns). These are meta-verification and introspection, not angle generation. angle-mort generates angles, it does neither meta-verification nor tooled introspection.

Generation rules

  • Volume 3 to 7-8 angles, depending on the richness of the subject. If it's forced, stop — a poor subject gives 3 good angles, not 8 mediocre ones. Better short and sharp than filler. If you sense a broader vein, propose to bring out others, don't stack them by default.

  • Imposed but flexible diversity of moves : aim for at least 3 distinct moves per run. Never 5 variations of the same reversal disguised as different angles — that would be a single blind spot repeated, not a true 360° view. Exception: if one move really dominates the subject, let it dominate, don't cap it artificially.

  • Raw output. No systematic stylistic polishing: we want material fast, not finished prose. The retained angle will be worked on later.

  • Anti-generic filter (non-negotiable). The #1 failure mode of a divergence engine is the angle that is deep in surface but hollow: “what if the real problem were something else?” Apply the replacement test to each angle: if I replace the subject with any other one and the angle still holds, it's slop — throw it away. Each angle must bite into a detail / a lever / a specific fact of THIS subject, named in the sentence. On the way out, go back over the list: cuts rather than dilutes (4 angles that bite are worth more than 8 of which half are interchangeable). No punchy but semantically empty formula, no unsourced psychologizing.

    slop (fits any topic → killed) sharp (bites into THIS topic)
    "What if the real problem was your relationship to the topic?" "What if the session were decided the night before, held ready, not in the morning?"
    "What if you changed your perspective on this?" "As seen by the skeptic: 1 session/week, does that really show visibly to the naked eye in 3 months?"

Output format

One line per angle, in this order:

- [Move] ⚡ « L'angle reformulé en une phrase. » `[tag]`
  → ce que ça ouvre (15 mots max, le potentiel brut, pas le cadrage narratif)

(The ⚡ is optional — see below. Most lines don't have one.)

The double tag states the epistemic status of the angle:

  • [à vérifier] — the angle rests on a fact, a figure, a factual claim that must be sourced before using it. (e.g., "the electric car costs 40% more" → to be confirmed.)
  • [spéculatif] — the angle is not a matter of fact but a bet on a point of view, a hypothesis. Nothing to source, but to be owned as such, not as a truth.

The line "→ what it opens up" gives the raw potential of the angle in one stroke, never the narrative framing. 15 words, cut beyond that.

The ⚡ marker (optional, 1 angle, 2 max) signals the most counterintuitive angle — the one that most reorganizes the topic if it holds. It is not the truest or the safest: just the strongest bet on asymmetry, the one that would change the game the most if it turned out to be true. It keeps its tag [à vérifier]/[spéculatif] like the others. ⚡ validates nothing — it is a bet on strength, not a verdict. Verification remains the next stage (up to you, downstream). It just gives a center of gravity to an otherwise flat list.

Example (topic: "I can't get back into exercise")

- [Inversion de direction] ⚡ « Et si je suivais une règle bête — sortir 2 min, le reste est optionnel ? » `[spéculatif]`
  → la barre devient impossible à rater ; l'effort réel suit tout seul.

- [Décalage temporel] « Et si la séance se décidait la veille au soir, tenue prête, pas le matin ? » `[spéculatif]`
  → la décision se prend avant l'instant de fatigue, pas pendant.

- [Test du clou] « Ai-je besoin de "faire du sport", ou juste de monter les escaliers au boulot ? » `[spéculatif]`
  → requalifie l'objectif sur un geste déjà dans ma journée, abaisse la barre.

- [Changement d'échelle] « Sur une année, une seule séance/semaine suffit-elle à voir un effet réel ? » `[à vérifier]`
  → si oui, la barre minimale s'effondre ; demande une donnée santé à sourcer.

  ~~[Inversion de direction] « Et si le vrai problème n'était pas la motivation mais autre chose ? »~~
  → tué (test du remplacement : tient sur n'importe quel sujet, ne mord sur rien).

After the list, suggest: "The ⚡ is the one I'd dig into first — do you take that one or another?" On the chosen angle, that's where you develop (what you used to output as an option, but on a single one, not on six).

Persistence

  • Ephemeral by default. The output stays in the chat, zero files. Most runs are disposable (quick thinking, personal decision) — we pollute nothing.
  • To keep only if the angle matters. If an angle deserves to be kept (a work topic, an important decision), copy it wherever you want — a note, a project doc.
  • Verification is a separate stage. An angle [à vérifier] is not ready as is: it is a fact to be sourced before you use it. Up to you, or to a dedicated verification tool.

angle-mort is brought to you by La MasterClass IA, the channel that gives you a concrete advantage with AI, without hype and without jargon. New video 2 to 3 times a week. → YouTube: @lamasterclassia · Going further with AI: masterclassia.substack.com

If this skill made you see an angle you were missing, say so in the comments under the video. That's what allows me to make more of them, and to offer them to you.

Reference

  • references/moves.md — the 8 moves in detail (5 core + 3 supporting): sub-variants, lead-in questions, examples.