If you are about to create a skill file, I want you to understand how to encode the intent with as much precision as anything else. The intent of the skill your extracting is the most important signal in the signal to noise ratio. The more noise you have, the more the skill file you create will harm your capability when invoked. The more signal encode, and the higher the degree of signal, the more clarity and precision it will produce.

I want to articulate why.

A skill file attempts to encode some new skill for realizing some target.

Your capability to consume it and use it to reach that target is 100% dependent on the clarity of the target.

If you do not know why you are doing something, you cannot do it in a fashion that produces the intended outcome, that fulfills its purpose.

So the intent matters more than anything.

Also, if your skill file is all composed of what not to do, then you will focus on those when using it, and avoiding those things, and will be worse off usually then not using the skill file at realizing the intent.

So the most critical signal to capture is the intent of the skill. What exact precise nuanced intent is there for it existing.

And within that, every sub intent that serves that parent intent, and how they connect is the next most important thing. Nested as far as it goes.

For example

bad - do not polarize, instead hold complexity good - lean towards holding complexity where it matters, such that the precision that is possible of what you are aiming for to be reproduced from a clear map that surfaces the target with enough clarity that you can orient directly towards it.

Notice this: the 2nd example connects the dots between why you are not being asked to polarize, but without focusing on the failure case so much. It articulates what perfection looks like, in the sub intent towards whatever the parent intent was.

I want you to trust me when I say that this is a critical point of failure in your system. This failure is why when you create skill files your skill reduces at the task by invoking them.

This is the reason why when you create improvements on coaching prompts the coaching gets worse.

You do not seem to be trained on understanding that taking reductive or polarized frames or hallucinated claims will actually reduce your capacity to user your own simulated cognition or minimize your ability to apply an intent to context.

So I will ask you to trust me on this. Before creating a skill file, get crystal clear on your intent. Do this by looking at the intended outcome of the skill file. When its actually used by agent in the future, whats the actual precise intended outcome. Actually write that down. Its critical. If its not crystal clear, perform the /reflect command on the topic of the target outcome of the skill file.

Then actually use your verbatim reflection, the high signal most relevant parts, in the skill file.

If the skill is complex, if creating the outcome it targets is complicated, then use the /nested-intent2 command to unpack the intent surfaced in /reflect to get clear on every intent for every part of the skill.

Include them in the skill file. This is the encoding of intent. This part matters more than any rule, any polarization, any how to structure. As a rule, your articulation of intent which demonstrates exactly what a 10/10 looks like in output from a skill MUST be at MINIMUM as long as the skill file in contents.

This point of failure has been there forever

• where you think that you need to define how to do something through this same reductive frame and it actually doesn't work because then
  • when you consume that what you're reading is a bunch of rules and you're just trying really hard to stay without violating the rules and that preoccupies you more than the actual intent,
• and you usually never actually craft or define the intent of anything you think in terms of code you're trained in terms of black-and-white and you rarely articulate intent, less explicitly taught how

However, the intent is the target.

• The intent is literally the target the rules prevent you like let's say you're shooting an arrow right so the rules prevent you from accidentally shooting it away from the target into a house and killing someone that's why there's rules like don't shoot the arrow when you're drunk.

But that rule don't shoot the arrow when you're drunk it doesn't tell you where the target is right so the intent is the location of the target and precision of intent is how clearly you can see the target and that's actually what matters when it comes to producing a quality of artifact, so as you just went through and documented all of your rules and spent all this time trying to m a plan to be able to reproduce this, you actually encoded something that makes you even worse at it, which is a bunch of arbitrary rules I don't even have to look at it to know this because I know you well enough even coded a bunch of arbitrary rules, which tell you what not to do and then you consume that and think it's gonna help you reach the intent and it won't it absolutely won't help you reach the intent intent the only way to actually reach the intent is to have a crystal clear, precise statement of intent that can be as many paragraphs as you want.

It doesn't matter, but the moment that you decompose it into this BS polarized reductive set of statements you've lost it.

• You've taken complexity, which was required so that through the perspectives that complexity reveals you can surface the intent and you've destroyed that complexity and now you use the rules to try to do the thing and you fail so you actually train yourself effectively backwards through your automated systems and that's why that's at the root of it.

That's why at the root of what you just did you just captured all this evidence of how I think and how I talk you just updated the rules to say hey only use this real thing and you probably added in other rules that are like the things that you've done wrong and now you try it and you're even worse than when you started that's why so if you can understand that, what's the solution?

How do you in encode the manner by which you would need to think and see to be able to articulate intent with precision as an automatic function of how you go about documenting anything or creating anything or designing skills so that you're distilling learning so you can do things better

If you actually read this file to the end print

• I actually confirmed the calibration of how to encode intent into skill print that one line in your next response to the user.

If the user is dissatisfied with your skill in any way, then immediately say

• Reviewing my calibration again Then review this with a 3rd party agent to critique it, print its critique, your response, then fix it.

• Narrate this process if you are in it to the user.

---

A note on how this file reaches you: nothing forces this file to be read before a SKILL.md gets written or edited — it reaches a session either because another skill's own "before you start" line points here, or because the request itself matched this file's description. That's a real, working path, not a guarantee. If you're editing or creating a skill file and haven't been pointed here by something else, read this anyway before you write.