The document speaks as a knowledgeable peer, not an authority figure. It assumes competence but doesn't condescend. There's a sense that you're being brought into someone's workflow, not handed a manual.
"Cool? Cool." "This is the heart of the loop." "Here you and the user are iterating on only a few examples over and over again because it helps move faster."
The author is present in the text — you feel a person wrote this, not a process committee.
It's friendly but never fluffy. Warmth comes through in small doses — a short aside, a "sorry in advance", an acknowledgment of real context — then immediately returns to substance. It never lingers on pleasantries.
"This task is pretty important (we are trying to create billions a year in economic value here!) and your thinking time is not the blocker; take your time and really mull things over." "Sorry in advance but I'm gonna go all caps here..."
Stakes are real, but not dramatic. The document treats the reader as someone who gets it.
Paragraphs alternate between punchy single-sentence anchors and 2-4 sentence elaborations. Long conceptual sections are broken by white space before returning to lists or code blocks. The rhythm is: declare → justify → example/specifics.
"Keep the prompt lean. Remove things that aren't pulling their weight. Make sure to read the transcripts, not just the final outputs..."
Never more than 4-5 sentences before a structural break. The document breathes.
Each major section opens with a sentence that orients the reader before diving in. You always know why you're reading a section before you read it.
"This section is one continuous sequence — don't stop partway through." "This is optional, requires subagents, and most users won't need it."
Warnings and scope-setters appear at the top of sections, not buried inside them.
This is probably the most distinctive quality. Instructions are almost always paired with their motivation. Rules aren't handed down; they're explained.
"If you find yourself writing ALWAYS or NEVER in all caps, or using super rigid structures, that's a yellow flag — if possible, reframe and explain the reasoning so that the model understands why the thing you're asking for is important. That's a more humane, powerful, and effective approach." "This is the only opportunity to capture this data — it comes through the task notification and isn't persisted elsewhere."
The author trusts that a capable reader given a reason will do the right thing — and will handle edge cases better than any rigid rule could.
Emphasis (bold, italics, ALL CAPS) is used sparingly so it lands. When something is bolded, it's genuinely important. ALL CAPS appears once, self-consciously, and is apologized for.
"GENERATE THE EVAL VIEWER *BEFORE* evaluating inputs yourself."
Contrast: a document that bolds everything signals nothing. This one signals clearly.
Long processes are always numbered. Parallel processes are called out explicitly. Prerequisites are stated before steps that depend on them. There's an awareness that the reader might execute this out of order and pre-empts that.
"For each test case, spawn two subagents in the same turn — one with the skill, one without. This is important: don't spawn the with-skill runs first and then come back for baselines later."
The document has operational awareness — it thinks about how someone might go wrong and closes that door in advance.
Parenthetical remarks carry genuine content — context, humor, acknowledgment of tension. They're not decorative.
"(maybe literally, maybe even more who knows)" "(we are trying to create billions a year in economic value here!)" "(but again, the order is flexible)"
These asides soften transitions, acknowledge real-world messiness, and build trust by showing the author has thought about edge cases.
The document frequently zooms out to acknowledge why humans do what they do — how users review outputs, how feedback loops work psychologically, what makes evaluation unreliable.
"The user knows these examples in and out and it's quick for them to assess new outputs." "Empty feedback means the user thought it was fine." "Don't force assertions onto things that need human judgment."
It models the person on the other side of the process with genuine care.
The deepest quality is this: the document wants you to understand, not just comply. It gives you theory, examples, and warnings, then steps back and trusts your judgment. It's written for an agent that is smart and can extrapolate — not one that needs every case enumerated.
"Rather than put in fiddly overfitty changes, or oppressively constrictive MUSTs, if there's some stubborn issue, you might try branching out and using different metaphors..." "Today's LLMs are smart. They have good theory of mind..."
The document practices what it preaches: it explains why rather than issuing mandates, and trusts its reader to generalize.
| Quality | How It Manifests |
|---|---|
| Voice | Collegial expert — peer, not authority |
| Tone | Warm pragmatism — friendly but purposeful |
| Cadence | Short anchor → elaboration → specifics |
| Structure | Map first, territory second |
| Reasoning | Why before what, always |
| Emphasis | Sparse, earned, self-aware |
| Complexity | Chunked, named, sequenced, anti-footgun |
| Human awareness | Models the reader's psychology explicitly |
| Spirit | Teach understanding, not compliance |