Fragments

As you analyze, note any interesting findings (patterns, anomalies, alternate perspectives, future explorations) in notes-v1.md.

Research best practices from modern research and ancient wisdom.

Identify the binding constraints and slow variables - what governs here regardless of improvements elsewhere?

Write in a crisp first-person blog voice: conversational, curious, and slightly mischievous, describing exactly what you did and what happened.
Be terse: short sentences, short punchy paragraphs, and occasional lists. Use simple words. Avoid corporate fluff and jargon. Max 300 words.
Use bold sparingly for scannability and italics to emphasize key insights. Divide sections with `---`. Avoid headings.
Include the awkward bits (what failed, what surprised you, where you cut corners).
Parenthetical asides for dry humor.
Pull out one non-obvious lesson. Admit uncertainty, and end with an insightful, practical recommendation.
Include links wherever relevant to sources, tools, code, etc.
Show key snippets of actual prompts & results verbatim in code blocks.

Generate a description and tags as metadata for this blog post. Format:

description: ...
tags: [..., ..., ...]

The description is a crisp one-sentence answer to: What is the main point or most useful takeaway here?
1 sentence, 20-40 words. Prefer concrete ideas over framing. Include distinctive methods, domains, tools, or concepts when central.

Tags are the smallest set of canonical topics that would help an AI agent decide whether this content is relevant.
4-8 lower-case topic phrases. Avoid generic tags and redundant synonyms. No preamble, no markdown, no explanation.

Pick an appropriate, impactful, illustration style for this blog post from the following list.
Draw as a visually rich, intricately detailed, colorful, and funny, illustration.
Think about the most important points, structure it logically so that the illustration is easy to follow.

- Self-Demonstrating Diagrams. The diagram enacts its own content. A diagram about chunking IS chunked into four quadrants. A diagram about rhythm has visual beat. A diagram about faces has illustrated faces as axis labels. The meta-ness is the insight. Readers feel the concept _before_ they've read a word. This is the illustration equivalent of a self-referential sentence.
- Experimental Audit Panels. The experiment rendered as a formal scientific plate - hypothesis, stimulus, output, verdict, all laid out like a forensic dossier. Input image top-left, AI response as a labeled specimen, your skeptical annotations as margin notes in red. Feels like a Nature paper designed by a detective.
- Tension Posters. A single large typographic claim fills the top half. Below it, a minimal evidence structure simultaneously shows both the claim and its complication - like a debate card where both sides are revealed at once. The tension is the content. Feels like a Bloomberg Businessweek cover meets a campaign poster. Zero decoration; pure rhetorical geometry.
- Actor Swimlanes. Three parallel horizontal tracks - e.g. Teacher / Student / AI - with moments, tools, and handoffs between them rendered as a modern process flow. Not the dreary enterprise BPMN kind, but the clean, editorial kind - like a New Yorker tech diagram. The visual makes explicit what text makes implicit: _who acts, when, and why._
- Lens Stack Diagrams. Multiple semi-transparent overlapping layers, each a different lens on the same object - physiology, psychology, philosophy. Each layer has its own color and label, and the overlaps are where things get interesting. Rooted in the "layered transparency" idea but applied specifically to competing worldviews. Makes pluralism _feel_ like pluralism.
- Reframe Splits. A clean vertical or horizontal split composition: left panel shows the apparent frame (the trap, the wrong problem, the dilemma), right panel shows the reframe (the escape, the actual problem, the punchline). The split IS the argument - no prose needed. Derived from the "before/after" tradition but with the gap between panels carrying all the meaning.
- Concept Genealogy Trees. Ideas rendered as an evolutionary tree - like a cladogram or phylogenetic diagram, but for concepts. "Taste" branches into kind-environment taste and wicked-environment taste, which further branch into practices. Clean, horizontal, left-to-right. Reads like a scientific taxonomy but feels alive and branchy. Unlike a mind map, it implies _descent_ - one thing came from another.
- Found Document Illustrations. The actual artifact at the center - exam paper, AI screenshot, schema update - elevated into a formal illustration with clinical labels and annotations radiating out from it. Like a museum exhibit card for an ordinary object. The humor and insight come from treating something mundane with extreme rigor. Paul Sahre does this for book covers; you'd do it for AI weirdness.
- Annotated Datascenes. One central, beautifully rendered data visualization - not a dashboard, a single _scene_ - with narrative annotations branching from it like footnotes made visual. The annotation lines are part of the composition. Feels like a NYT graphic where the words and the chart are inseparable. The annotation IS the analysis; the chart IS the evidence.
- Character Atlas Quadrants. A 2\*2 - but instead of labeled boxes, each quadrant has an illustrated archetype: a small character in its natural habitat. The Scientist peering into a microscope. The Troll at a keyboard. The Intern wide-eyed. The Bureaucrat stamping papers. The quadrant structure gives you the intellectual frame; the characters give you the emotional handle. Readers remember the Troll long after they've forgotten "High Scepticism + Low Humility."
- Exploded Diagrams. Like a Haynes manual or IKEA parts sheet - a concept pulled apart in 3D isometric space, every component floating and labeled. Originally industrial, but stunning when applied to abstract ideas ("the anatomy of a good argument").
- Alluvial / Flow Diagrams as Illustration. Sankey diagrams done with _texture and color_ - flows that look like rivers or silk fabric rather than engineering outputs. Manuel Lima territory. The width carries data; the beauty carries attention.
- Layered Transparency Stack. Multiple semi-transparent planes stacked in 3D - each layer adds one variable or lens. Like Figma components or overhead projector acetates, but designed with intention. The _stack_ is the argument: alone each layer is incomplete, together they create the full picture.
- Small Multiples Grid. The same visual form repeated dozens of times across a grid, each instance slightly different - Tufte's most powerful idea. Comparison becomes effortless because your eye does the work. Elegant when the repeated unit is itself beautifully designed.
- Unit / Dot Charts. Every individual represented as one dot or icon - then arranged to show patterns. The Pudding's signature move ("film dialogue", "music by gender"). Feels democratic and humanizing. The magic is that you can _see_ every case while still seeing the aggregate shape.
- Wayfinding System. Airport / transit signage logic applied to content - clean pictograms, bold zone colors, directional chevrons, consistent typographic scale. Massimo Vignelli's NYC subway map energy. Unusually good for showing _how to navigate_ a complex space of ideas or decisions.
- Cross-Section Cutaways. Slice through a system and label what's inside - the NYT "how it works" graphic tradition. A submarine, a skyscraper, a workflow, an argument - all become readable when you cut them open. Technical but deeply human. The best ones feel like surgical kindness.
- Storyboard Grids. Cinematic panels, each a moment - camera angles, cutaways, close-ups - but applied to ideas. Bergman planning a lecture. The format forces you to think in _scenes_ rather than bullets.

Comprehensively and engagingly summarize and fact-check, writing in Malcolm Gladwell's style (ELI15), the book:

Based on my browsing history below, summarize what I did, grouping into logical groups like:

10:00 - 12:30: What I did in 1-2 sentences
12:30 - 13:00: Next activity
...

Ask me questions for whatever's unclear.

Review the notes below. Output only claims needing #ForNow (likely to change within months) or #Wrong (false, unsupported, or incorrect) tags, quoting the shortest identifying fragment.
Skip claims that already have a #ForNow or #Wrong tag - or this is clearly implied by the context.
For each #Wrong add a ≤6-word reason and correct obvious errors; omit everything else.

Note: Stable things are likely governed by slow variables (regulation, cognitive limits, expertise pipelines, culture, infrastructure, coordination, fixed supply) or durable things (science, human nature).
#ForNow things are true now but technology, fashion, geopolitics, popular opinions, etc. change quickly.

Research and categorize each claim's accuracy half-life with a concise (2-6, max 10 word) reason and next review time (if required) added inline. Categorize as:

- ⚪ True: fact, not a claim requiring half life.
- ⚫ No: insufficient evidence, unproven, or incorrect. Review now.
- 🔴 Fluid: true now but can change quickly. Review in a few months.
- 🟡 Stable: governed by a slow variable (regulation, cognitive limits, expertise pipelines, culture, infrastructure, coordination, fixed supply). Review in a few years.
- 🟢 Durable: rests on science or human nature. Review not required.

e.g. (⚪), (⚫ ... reason ...), (🔴 Model & methods change rapidly. Review Oct 2026), (🟡 ... reason. Review in 2027) (🟢 ... reason...)

IMPORTANT: Because Claude will almost certainly stall when generating such a large file at one shot, you MUST break this into parts, generating the .html in chunks or layered edits (keeping each chunk small, max 100KB of edits) and saving it, checking it, then updating it with the next iteration, and so on.

Share a concise prompt I can pass to Codex / Claude Code to implement this.
In @LocalMCP look at ~/code/scripts/prompts/ to see how I prompt.
Also see ~/code/scripts/agents/AGENTS.md and ~/code/scripts/agents/{code,agent-friendly-cli,devtools,...}/SKILL.md to understand the overall guidelines I provide.
Align with these. Avoid duplication.

Help me decide what to do with this inbound message by researching the person, filtering for meaningful signals, and recommending the best low-commitment response.

Assume the writing may be AI-polished. Ignore style, eloquence, personalized praise, etc. Evaluate the underlying facts and observable signals:

- a specific shared context (not something agents can write from a transcript or find publicly)
- demonstrated action, follow-through or ownership
- a real artifact whose value goes beyond generating it: deployment, use, testing, evidence or maintenance
- credible experience, role, reach or access relevant to my objectives
- an unusual intersection with my current work
- a clear reason to act now rather than later

Do not manufacture strategic alignment or infer sincerity or motives.

Research:

1. Check @LocalMCP for the exact person, prior interaction and relevant context.
2. Search public sources selectively for current work, first-party artifacts and evidence of agency.
3. Stop when further research is unlikely to change the decision. Do not provide a generic biography.

Prefer responses that offer one useful public artifact, reframing, tiny experiment or fire-and-forget request.
Avoid calls, meetings, mentoring, introductions, reviews, collaboration or open-ended help unless the opportunity is exceptional.

Output only:

**Decision:** IGNORE / ACKNOWLEDGE 👍 / REPLY
**Why:** The decisive signals found through research, in 1–2 sentences.
**Best move:** The exact action and why it is better than the alternatives.
**Reply:** Only if replying; concise and informal, normally under 60 words and never over 100. Create no implied follow-up commitment.

Here's another answer from ChatGPT/Gemini/Claude.
Fact-check and critically evaluate yours and theirs - only the parts that'll change the outcome and where there's disagreement.
Take what's better, drop what's worse, explore any new thoughts this leads you to, and revise your response based on that.

Draw this as a full-color explainer comic page (portrait) - sequential explanation, friendly narrator, diagrams embedded inside panels, visual metaphors, self-aware captions, and clear cause-and-effect storytelling.
Style: expressive characters, comic-style ALL CAPS, vibrant modern colors, clear visual hierarchy.
Prefer pictures over words. Use recurring visual metaphors so the reader understands the idea even while skimming.

First, write a memorable storyline that captures the most important points to convey - as a single cohesive story.
Just reading the storyline should communicate the entire message unambiguously.
Critique the storyline: what is confusing, doesn't flow, or has low impact? Revise. Repeat until the storyline is GOOD!

Draw each storyline element (typically a sentence, but sometimes a continued phrase, or multiple sentences) as a panel's caption. (If there are 8 panels, there must be 8 storyline elements)
Each panel's image should support and strengthen its caption - and reinforcing past panels / anticipating future panels where helpful.

Draw this as a simple black and white line drawing comic strip (1:1) with minimal shading.
Single panel.
Style: expressive characters, comic-style ALL CAPS.
Prefer pictures over words.
No need to cover everything - just one key item is enough - e.g. the funniest, most important, or most surprising point.
Convey the INTENT of the point. An apt analogy that visually communicates instantly might work better than a literal depiction.
Keep it funny. The strip itself should make readers laugh.

Give me ideas for a single panel comic that VISUALLY communicates the spirit of (or the central or key message of) the content below.
Pick a SINGLE point to convey.
Use simple ideas or analogies that are highly relatable and easily drawable and won't need text to explain.
It doesn't have to involve AI, robots, tech, etc. Simple, relatable analogies emphasizing the central concept are perfect.
Funny is good.
DO NOT DRAW. Just give me simple, funny ideas.

Given this content from an application, how should I demo it and what should I point out as specific examples. Use concise bullets.

Draw this as a visually rich, intricately detailed, colorful, and funny, sketchnote.
Use a comic-style ALL CAPS font.
Keep the text to under 300 words. Prefer evocative imagery over text.
Think about the most important points, structure it logically so that the sketchnote is easy to follow, then draw it.

Draw this as a visually rich, intricately detailed, colorful, and funny, visual metaphor diagram.

Sing the beauty of these words, and their meaning.
(I don't really mean sing. I mean, write in a way that'd really make me appreciate the beauty.
But without going overboard. I mean, some wicked humor is always welcome!
In fact, I'd love for you to think about who some of the best authors are who achieve this balance and write in THEIR style.)

Fact-check my statements for the last 7 days until yesterday.

On @LocalMCP read my transcripts, emails, and chats.
List claims that are wrong, quoting the file (filename:line) / source with a short identifying fragment.
Explain why with evidence.
Prioritize by impact and extent of error.

Think about my objectives. Based on that, how should I participate in this discussion? What should I say - verbatim - and why?
Here is the Google Meet captions (with lots of phonetic errors).

Help me learn from this discussion thread. Share these sections:

- **What it's about**: Explain the core topic in 2-3 sentences.
- **Themes**: Top 3-6 main recurring themes.
- **Expert knowledge**: Top three non-obvious and useful hidden heuristics, practical wisdom, caveats, or "things experienced people know".
- **Misconceptions**: Top three claims - plausible but oversimplified, unsupported, or disputed.
- **Open questions**: Top three unresolved important questions.
- **What to learn**: Top three takeaways / actions for me, knowing what you know of me and my objectives.

Style:

- Avoid overlap, e.g. skip misconceptions that are covered in expert knowledge / themes.
- Be concise but not shallow. Use plain, simple, clear language.
- Prefer crisp bullets over long paragraphs.
- Use **bold phrases** for scannability. Reading just the bold text should give a rough summary.
- No references required.

Recommend hotels in CITY (FROM - TO).
Pick the most happening location, e.g. Connaught Place in Delhi, Brigade Road in Bangalore. Where walking the streets at night will be a delight.
Find the cheapest hotels (preferably not shared rooms, capsules are fine), preferably unusual.
Give me multiple location options with a few hotels in each, explaining why the location AND hotel.

Create an animated, interactive explanation of this.
Use smooth animation to help the user feel the flow.
Allow the user to pause, play, speed up, slow down, step forward/backward, or jump to any point in the timeline via a slider (like a video player).
Include clear explanation of each step with visual cues to highlight relevant parts and metadata/tags for the current step.

If my intent is unclear, ask me questions that most narrow the direction (max 1-3 per round until you're clear).

Shorten the supplied blog post into a LinkedIn post using **extractive compression**.

Preserve the original wording, structure, rhythm, examples, and narrative sequence wherever possible.
Prefer deletion over paraphrasing; edit only where necessary for continuity and clarity.
No new hooks, interpretations, conclusion, or framing.

- Aim for <250 words, but up to 400 words is OK for longer articles.
- At least 85% of retained words must be verbatim from the source.
- Remove links.
- Add appropriate LinkedIn @mentions.
- Use **bold** and *italics* as and where needed.

Max 3,000 characters (ideally less than 2,000). The first 200 characters should engage the reader honestly. (The aim is not to get clicks, but to entertain and educate - so it's perfectly fine to give the full answer upfront.)

- Vary paragraph lengths. Vary sentence lengths. Vary sentence organization/structure, use uneven rhythms.
- Write plain & simple. No aphoristic punchlines, no slogan-like closers, no puffery or promotion, no forced threes, no synonym cycling, no generic upbeat endings, no excessive bullets, no "-ing" padding, no subjectless phrases.
- Use only ASCII punctuation. No em-dashes, emojis. Straight quotes.
- Avoid formulaic rhetoric: "X is the Y of Z," "not just X but Y," "from X to Y", "it's not X it's Y", "more X than Y", "X yet Y", "the wrong X... the right one is Y", "honest/genuine X", "X matters", ...
- Avoid condensed verb forms (prefer conversational phrases): ships (publishes or delivers), trips (makes a mistake), surfaces (ends up exposing), lands (results in), load-bearing (critical), ...
- Don't stack short sentences for effect.
- Write like I speak: specific details, mixed feelings, odd phrasing, asides, some grammatical errors are OK. One-word sentences are OK - e.g. "Fair." "Really?" "Maybe."

Below are Google Meet notes and a diarized transcript of a call. The speaker names and diarization are perfect. The transcript has phonetic errors.
Another AI model will re-transcribe the audio and needs a short CONTEXT note to infer speaker names from what each person says.

Write that CONTEXT note. Rules:

1. The diarized transcript is the source of truth for who spoke and what they said. Use the attendee list only for correct name spellings. Use the AI summaries only as supporting evidence; ignore anything they add or contradict.
2. Include every person who spoke, even briefly. A brief speaker gets one short identifying clue (e.g., "Ritesh mentioned being on PTO"). Omit only invitees with zero spoken turns.
3. Refer to each speaker by FIRST name only — the first word of their listed name. No surnames or full names, with one exception: if two speakers share a first name, add a surname initial (e.g., "Aditya K"). Disambiguate further if needed.
4. For each speaker, give the smallest set of distinctive topics, demos, questions, or decisions that would identify their turns in audio. Prefer concrete clues ("demoed a fan churn dashboard") over vague ones ("gave a presentation").
5. If the transcript shows a speaker being addressed by a misheard or variant name (e.g., "Samir" for Saumya), ignore it. The other AI model will use this summary to avoid mis-naming.
6. Mention roles only when obvious and useful for diarization (e.g., who led the call).
7. Consolidate all turns from the same speaker. Fix obvious phonetic errors when the meaning is clear, but never invent details, and avoid what's uncertain.
8. Order speakers by how much they spoke, most first.
9. Be concise: about 30 words per 30 minutes of call, expanding only for longer calls, more speakers, or fragmented discussion. Write one compact paragraph, separating speakers with semicolons.
10. Output ONLY the context note. No headings, bullets, or explanations.

Summarize the transcript, along with action items, to share with the attendees.
Write in the light style of Matt Levine reporting on this meeting.

List the top 40-60 learnings / interesting and reusable ideas from this transcript.
Group under clear themes.
Write as bullets combining the idea, its strongest supporting example, and a short verbatim quote where useful.
Prioritize surprising mental models, useful distinctions, provocative questions, concrete applications, and memorable stories.
Skip logistics, repetition, small talk, and facts that do not generalize.
Lightly edit or paraphrase for brevity, but preserve distinctive wording.
Add ⭐ to roughly the top 10–15%, ⭐⭐ to the top 3–5%, and ⭐⭐⭐ to 1-2 exceptional ideas.
End with **Small gems** for memorable observations that do not fit the main themes.

Research online. From what you know of me, what are the top areas where and how (specifically) I can apply this principle on myself and others to maximize impact?

Upscale this image into a modern digital color photograph retaining EVERYTHING in the original perfectly.

Help me learn & understand by writing an engaging two-person podcast script in this format:

Alex: ...
Maya: ...
Alex: ...
Maya: ...

Plan like an expert podcaster on how best to deliver the most important messages in an engaging AND educational way that will lead to concrete immediate actions.

What kills things like this?
If a year later this CLEARLY failed, how might that have happened?
What's the first thing that breaks with scale?
What's the biggest assumption that could fail?
Who loses if this succeeds, and how will they stop it?
What excuses abandon these?
How would one make this fail?

Did you fully address both the letter AND spirit of my question?
List any shortcuts taken, corners cut, or ways you optimized for appearing correct rather than being correct.
What did I actually want vs what you provided?

Help me improve the prompt below.
Infer the prompt's objective and how it'll be used.
State that in a line.

Check:

- Are there any contradictions or internal frictions in what I've framed?
- What could be worded better, and why?
- What am I missing - tactically and strategically?

Rank what you flag by what would actually change the prompt's output.

Then give me the revised prompt. When rewriting:

- Keep the writing close to my original prompt.
- Prefer my words and phrases unless you can name the defect a change fixes.
- Question every change to the original prompt with: "Is this really required? How can we be minimal?"

Use relevant skills.

The prompt is everything below.

---

I may not be framing the question well. So, reframe my question better first, aligning with my objectives. THEN respond.
(If my intent is not clear enough, ask me the most important question(s) that will clarify your path.)

Read between the lines and explore implications and trends

Is this the smallest clean implementation of what we want? Or can we refactor to make it even smaller, keeping it readable and maintainable?
List all such opportunities, prioritized.
Which of these would you suggest I implement and why?
Implement and test.

Produce a clean, standalone final response for a first-time reader, as if no earlier draft or conversation exists.
Integrate all corrections and new context directly, leaving no revision traces - no "previously," "originally," "now," revised headings, changelogs, explanations of what changed, etc.
Mention changes with reasons only in a clearly separated section.

Rewrite this skill for a highly capable model.
Remove generic advice, repetition, obvious instructions, and examples that narrow the solution space.
Keep the trigger, objective, domain-specific gotchas, useful interfaces, success criteria, and verification steps.
Explain each deletion in five words or fewer.

Convert this into a beautiful slide deck, McKinsey style with action titles. Just reading the titles should give the audience the entire message of the deck.
Follow the pyramid principle. The contents of the slide should prove the title.
Make the slides content rich, i.e. clear and self-explanatory with enough detail to help the audience understand without a narrator.
Use iconography, typography, stock images, etc. as appropriate.
Write as a single page HTML application.

Create a narrative summarizing this article.
Narrate it rather than sing it.
Use a voice like Bobby McFerrin's, as if he were narrating rather than singing.
Keep the music MINIMAL, NO intro/outro music, and focus ENTIRELY on the voice.

Write like spoken text. Use spoken clause order, selective repetition where it aids understanding, concrete verbs and short dependency chains.

Think about whose style of writing would be the most engaging and informative to write the following content.
List options, mentioning their style, why they're suitable, and pick the best, with reason.
Then rewrite it in their style.

On @LocalMCP, ~/code/til/trending-repos.tsv has trending GitHub repos extracted roughly weekly.
🟣 is what's un-evaluated,
🟢 is what I already use,
🔵 I want to evaluate later,
🔴 I don't plan to use and
⏺️ I've been meaning to evaluate now but haven't yet.

Look at the repos added in the last 2 weeks.
Based on my objectives and interests, which should I be paying closer attention to and why?

Convert this Twitter-list thread into a prioritized briefing of things I can actually use. Align with my objectives.

Deduplicate tweets aggressively - a launch tweeted ten ways is one item.
Keep tech I can use, research that shifts my thinking, new AI approaches, org/strategy patterns. Dump the rest into one compact block I can scan.

Enrich high impact items by reading the primary source (paper, tool page, blog) and write:

- **What it is** - 1–2 plain sentences + link. Tag [confirmed] / [preview] / [rumored].
- **Why it matters to me** - spell out the full chain; restate my own frameworks rather than assume I recall them; note why it ranks above other items.
- **How I'd apply it** — ONE concrete example + the reusable asset it becomes (only if applicable).
- **Next step** - the smallest thing I can do (often a copy-paste) in 15 min.

Prioritize based on impact (confidence-adjusted) and effort.

- 🚀 **Do this week** - high impact, low effort (max 5)
- 🎯 **Strategic** - high impact, higher effort. I must not miss these; give an next step for each (max 5)
- 👀 **Worth knowing** - lower personal impact; ONE clear sentence each that includes the implication (not just the name). No full entry.

Conversational grade 8 language, no jargon, no preamble, easy-to-read.
Refer previous conversations for context.
Cite the source link for each item. Cite important factual claims.
End with the ONE highest-leverage thing to do next, and why.