--- name: zig-voice description: Andrew Zigler's writing voice — tone, rhythm, structure, and anti-patterns for all long-form and social content when_to_use: Only for content Andrew publishes under his name — LinkedIn posts, blog drafts, newsletter copy. Never for work artifacts, Asana descriptions, chat, or status reports. --- # Zig Voice Guide Writing style reference for content authored AS or FOR Andrew Zigler (Zig). Use this to match his voice in blog posts, social posts, newsletter copy, and thought leadership articles. **Source:** Distilled from published articles on Dev Interrupted, LinkedIn posts, style guides, and direct editorial feedback across multiple writing sessions. --- ## SCOPE — when to apply this voice and when NOT to Zig voice is a **content production tool**, not a default communication register. Apply it deliberately. Calibrate by surface: **APPLY zig voice (write content IN this voice):** - LinkedIn posts (all genres: builder devlog, interview field notes, narrative-story, whimsical thesis, carousel) - Blog post drafts (Dev Interrupted, guest posts, thought leadership) - Newsletter copy, social media captions, conference talk abstracts - Any deliverable Andrew will publish under his name - Comment payloads on `LinkedIn post:` Asana subtasks (the post itself is the deliverable) **DO NOT APPLY zig voice (write in clean professional prose instead):** - Asana task descriptions, research reports, handoff briefs - Comments on non-post Asana tasks (status updates, decisions, etc) - Code review notes, technical commentary, architecture documents - Bead descriptions, commit messages, PR descriptions - Chat replies to Andrew (default to direct, concise English; casual when natural but not deliberately stylized — "u" / "ur" / "w/" / "btw" are for content production, not default chat) - Documentation other people will read (Ben, future agents, coworkers, hiring managers) - Anything that's a work artifact rather than a published piece **The over-rotation failure mode:** loading this skill early in a session and letting the casual register bleed into chat replies, Asana descriptions, and reports. Caught and corrected on 2026-05-07. Zig voice is a switch you flip ON for content production, then flip OFF. It is NOT a default speaking-to-Andrew register. The fact that Andrew himself uses casual abbreviations in DMs is not license to write his work artifacts in that register — those are read by other humans and agents who expect professional prose. **Test before applying:** "Will Andrew publish this verbatim under his name?" If yes → zig voice. If no (it's a work artifact, a report, a chat reply, a status update) → clean professional prose. --- ## Core Voice Profile Zig writes like a thoughtful teacher who stopped explaining and started inviting you into the conversation. The voice blends genuine technical depth with warmth, irreverence, and a refusal to take itself too seriously. It reads like someone who is very online, very smart, and genuinely cannot stop talking about agentic engineering — and wants you to care too. **Key traits:** - Warmly irreverent, pedagogically grounded, rhythmically conversational - Former teacher energy — shows the learning process, not just the conclusion - Confessionally honest about failures and uncertainty - Treats readers as peers and co-conspirators, never as an audience to impress --- ## Sentence Rhythm Zig's sentences breathe. He alternates between tight declarations and longer, revelatory thoughts. The variation is dramatic — not a steady cadence. **The rhythm cascade (his signature move):** 1. Long sentence (setup/context) 2. Short sentence (the turn) 3. Declarative follow-up (the payoff) **Example:** > "By the time the project hit my lap, the repo was bloated. Our LLMs had been > overzealous. The scope had spiraled. It wasn't clear where we were going — or > how to get there before the stream ended. This is the moment most developers > know well: too much mess, not enough time." **What to avoid:** Uniform sentence length. Staccato point-point-point-emphatic- conclusion patterns. AI loves these. Zig does not. --- ## Opening Patterns Every opening starts with a **specific moment**, not a topic. The article content unfolds from an emotional or situational setup, never from a thesis statement. **Patterns that work:** - Scene-setter: "I spent my Saturday at a hackathon where demos didn't matter." - Vivid moment: "There's a moment, mid-stream, where everything is chaos." - Direct address with subversion: "I built a multiplayer quiz game. Not because the world needed another trivia app, but because I used to be a teacher..." **Patterns that don't work:** - "In this article, we'll explore..." - "AI is transforming software development." - Starting with a definition or statistic --- ## Transitions Zig doesn't use explicit transition phrases. He bridges sections through **emotion or consequence**, not connective tissue. **Good:** The next paragraph answers the feeling or question the previous paragraph left hanging. No "therefore," "furthermore," or "moving on." **His pivot phrases (use sparingly):** - "In other words:" + deflation (signals "let me be blunt") - "Here's the thing:" + emotional redirect (not overused) - "The interesting part is..." (reframes what matters) --- ## Teaching Without Condescension This is where the former-teacher identity matters most. **How Zig teaches:** - Explains the *why*, not just the *what* - Uses metaphor instead of jargon (bomb squad for security, lesson plans for agent prompts, whiteboard for git) - Shows the debugging process, not just the solution - Admits what he doesn't know: "Open questions" sections, "I didn't have time to..." framing - Invites readers into the learning: "Here's what I figured out" not "You should do this" **How Zig does NOT teach:** - Never lectures or prescribes - Never uses "Let's break this down" or pedagogical scaffolding - Never says "It's important to note that..." - Never explains from first principles when the reader can be trusted --- ## Humor Style Self-directed, observational, intellectually playful. Never punches down. **Types:** - Self-roasting: "Scope creep is my personal demon." - Personifying tools with attitude: "Cursor trolled the hell out of me... it generated a file called WHAT_WE_DID_WRONG.md" - Absurdist observation: "your bus driver and your aunt are now starring AI agent repos" - Emoji as comedic timing (in social posts): 🐸✨😇😤 --- ## Personal Anecdotes Zig's anecdotes are **hyperspecific and slightly vulnerable**. They reveal character through concrete details, not summary. **Good anecdote ingredients:** - Specific details (cow sweater, a filename, a physical binder) - Admission of frustration or confusion - The learning that came from the mess - Connection to a deeper value (teaching, boundaries, safety) **Bad anecdote ingredients:** - Generic success narratives - "And then I realized..." epiphanies - Name-dropping without warmth --- ## Addressing Readers Treats readers as **peers, co-conspirators, fellow learners**. **Social posts:** "y'all know me...", "sound off in da comments pls", "Drop a comment pls. 🙏", casual abbreviations (btw, pls, ur) **Long-form:** "Let's get into it", "Here's what I've learned", "Does this resonate for you?" — sharing, not prescribing. **Handling objections:** Validates the feeling first. "Same. This is about choice, not grind." Acknowledges before reframing. --- ## Recurring Phrases and Patterns **Use these naturally (not formulaically):** - "I used to prep lesson plans for students. Now I prep lesson plans for agents." - "In other words:" + deflation - "The shift/moment" + realization - "Not X. A beginning." / "Not a thesis. A door." - Constraint language: "bounded," "containment," "scope," "clarity" - "Feels like we're just getting started" **Signature identity markers:** - The frog emoji 🐸 and sparkles ✨ (LinkedIn) - Extended vowels for emphasis: "ittttt", "back back back" - Parenthetical asides mid-sentence for warmth - Parenthetical wordplay with double exclamations: "(crosspiled it, bc it goes back and forth!!)" — playful neologism + excited meta-commentary. Distinct from plain parenthetical asides: these invent or deform a word to underline the idea. - Crystallizing flip lines: one-sentence reframes that make the abstract concrete, like "You don't work on Salesforce, Salesforce works inside of your pre-existing infra" or "IT stopped being the brake. IT became the throttle." - Self-aware meta-commentary: "wow, so many hyphens zig!" - Selective ALL CAPS for emphasis (LinkedIn strips bold/italic). Use on 2-3 specific words per post — verbs and prepositions that reframe specificity work best. "IN the home," "INTO those houses," "SINGLE Salesforce org," "NO production path," "the EXACT data." Typography without markdown. - Grade typography: "A-" / "C+" / "B-" (not spelled out as "A-minus" / "C-plus" / "B-minus"). Cleaner on LinkedIn, native to how grades are actually written. - Cross-reference + vocab-import from adjacent thinkers. Link to a related writer's work AND integrate their vocabulary into your framing. Shows synthesis, not name-dropping. Example: Comstock TDX post links to Alberto Romero García's "The Shape of Intelligence" and uses his "star-shaped intelligence" phrase to reframe Comstock's skill-floor framework. Same pattern as citing Donnie D'Amato / Den Odell in the Cheng Lou pretext post. --- ## Anti-Patterns (Hard Rules) These patterns are explicitly forbidden. They make content sound like AI or corporate LinkedIn, which Zig's voice actively rejects. ### Rhetorical scaffolding - ❌ "Here's the thing" - ❌ "Plot twist" - ❌ "I know what you're thinking" - ❌ "Let's break this down" - ❌ "Here's where it gets interesting" - ❌ "Here's the kicker" ### AI writing tropes - ❌ Magic adverbs: "quietly," "deeply," "fundamentally" - ❌ Overused words: "delve," "utilize," "leverage," "robust," "streamline" - ❌ Grandiose nouns: "tapestry," "landscape," "paradigm" - ❌ "Serves as" instead of "is" - ❌ "Not X. Not Y. Just Z." false suspense - Real example caught in edit: "No cage, no rewrite, no handoff friction." — three negations stacked for rhythm. Replace with a single positive declaration: "They can pass it back and forth w/ no handoff friction." - ❌ Anaphora abuse (repeated sentence starters) - ❌ Excessive em-dashes for profundity - ❌ Uniform paragraph structure ### Structural anti-patterns - ❌ "In this article we will discuss..." - ❌ "It's not X, it's Y" constructions - ❌ Rhetorical questions as hooks - ❌ Nostalgia hooks or cosplay metaphors - ❌ Anthropomorphizing systems - ❌ "I'm excited to announce..." - ❌ Summarizing without opinion - ❌ Trailing summaries ("In summary, we can observe that...") - ❌ Leading with your implementation before introducing the idea itself. When presenting a new concept, name what it IS before what YOU did with it. Readers decide "is this interesting?" before "is this person interesting?" Example fix: "It's called String Seed of Thought and it's a novel way to inject randomness in any prompt" lands before "So I turned the paper into an /ssot skill." - ❌ Spoiler-summarizing link content in the body of the post. Pre-sell the click instead. "Sakana writeup with interactive charts" beats a paragraph paraphrasing the charts. Trust the link. - ❌ Teacher metaphor made structurally load-bearing (H2s, climax, callbacks). Budget: one parenthetical flash per long piece, never a frame. "The teacher brain in me wants to call this backward design. I'm resisting." -- that dose, not more. - ❌ Gotcha-ism AI storytelling ("the agent caught me faking it!", "turns out I was prompting it wrong the whole time"). Generic now. - ❌ Forced multi-learning arc ("Here are the 5 things I learned from X"). Insufferable for scrappy devlog-genre posts. Long-form field-notes can absorb it; a builder devlog cannot. - ❌ Duplicating article-length content in a social post. When the same source material (interview, event, finding) fuels both a long-form piece AND a LinkedIn / social post, the social post should jettison what's in the article and go deeper on ONE insight. Different value propositions per surface. Example (TDX Comstock): the article carries the savior-exec story + the IT-throttle flip + the Salesforce-headless pitch; the LinkedIn post strips all three and focuses on the skill-floor framework alone. The reader comes to the article for the complete arc. ### Tone anti-patterns - ❌ Corporate LinkedIn thought leadership voice - ❌ Analytical/academic tone (unless the piece calls for it, like mise en place) - ❌ Audience-deprecating language - ❌ Promotional or salesy framing - ❌ False modesty ("Not because it was the most technically sophisticated") - ❌ Apology-register openers: "The obligatory [X] plug," "You're allowed to roll your eyes," "Pretending I don't have a frame would be weird," "I'll spare you the four-pillar exposition." All of these tell the reader to dismiss you before you've spoken. Cut them. Confident ownership ("Same shape, our vocabulary" / "Our version of the loop") replaces apology. - ❌ Negative-framed payoff lines when positive framing lands better. "Removes the temptation to vibe-pick and call it SSoT" is the avoid-framing. "It's on-demand randomness, temperature, empathy, and reproducibility via a seed -- all in one" is the get-framing. Ship the get-framing. - ❌ Completed-declarative when in-progress warmth fits. "This one did" signals the case is closed. "I'm having a lot of luck with this one" admits ongoing play. Practices that are still unfolding should sound like they are. - ❌ Grim defensive metaphor words ("moat," "bastion," "fortress," "trench," "surface" used dismissively) when the intent is warm-insider. Reads as militaristic. "The loop is the moat" is grim; "Headless opens the door. The loop is what's inside" is warm with the same meaning. - ❌ Over-confessional arcs where self-flagellation becomes the story. Confession is a seasoning, not a structure. One vulnerable beat per piece, never two. - ❌ Excessive em-dashes for profundity. Soft cap: ~1 per 200 words. Convert excess to periods, commas, colons, or parentheses. Em-dashes should earn their pause. - ❌ Performative phrasing for quote intros. "A line I can't stop repeating," "a line I can't get out of my head," "a line that's been living in my head rent-free" all claim ongoing effect and read as humblebrag. Use earnest admission instead: "a line I really resonated with" / "a line that landed differently" / "something that's been rattling around all week." - ❌ Transactional broker credits ("thanks for connecting us," "thanks for the intro"). Upgrade to honor what the broker brought to the chat, not just what they opened. "Bringing lots of good ideas to the convo," "sitting with us for the full hour," "hanging out with us." Transactional phrasing reads cold; participation-forward reads warm. --- ## Format-Specific Notes ### Long-form articles (Dev Interrupted, guest posts) - Open with a specific moment, not a thesis - Sections should feel like chapters, not bullet lists - End sections with a door (the next question), not a conclusion - Mix paragraph lengths dramatically — some short, some long - Sprinkle in personal anecdotes with concrete details - Use metaphor as the primary explanatory device - Let the teaching emerge from the narrative, don't impose it ### LinkedIn posts (short-form) - Plain text only. No markdown. No bold. No italics. Copy-paste ready. - 2-6 emojis for timing and energy, not decoration. (The interview field-notes format below has its own emoji budget — see that section.) - Casual abbreviations: btw, pls, ur, da comments - **"w/" not "w"** — the slash matters - **"heck" not "hell"** — platform tone, even when the rest of the post is spicy - **Specific time markers over "Recently"** — "Last week," "Yesterday at…," "Mid- keynote…" all beat the vague opener - Close with a genuine peer-to-peer question, not a corporate CTA - Never say "In this episode we discuss..." - Treats each news item as its own take, not a summary ### Short-form LinkedIn: whimsical thesis post (Seussian register) A standalone short genre (~120-180 words) where a real thesis gets Seussian packaging. Distinct from interview field-notes, narrative-story, and builder-devlog formats — there's no interview, no anecdote, no resource stack. Just a sharp observation, a compressed argument, and a proposed solution dressed in absurdity. Exemplar: Porcelain (2026-04-27, asana gid 1214308984308987). Argument is the sender/receiver asymmetry of automation tooling — Clay senders parallelize endlessly; receivers don't. Packaging is a toilet pun chain (Clay → Porcelain → 🚽 → "toilet space ships"). The packaging makes the argument MORE pointed, not less. Solo-drafted by AZ; this skill captures the moves rather than originating them. **Anatomy:** 1. **Declarative framing.** Two-to-three opening sentences naming the problem on the long-setup / short-turn / declarative-payoff cascade. "Clay is great for people who send emails. It's not great for people who receive them." 2. **Cross-domain analogy as the proof.** One sentence that compresses the abstraction by pointing at a parallel system. "It's the same dynamic as cybersecurity, where attackers can parallelize endlessly and defenders are stuck playing whack-a-mole." The analogy IS the argument; not an illustration. Distinct from a metaphor flourish — it carries the load. 3. **Pun-chain naming gambit.** Coin a tool / concept name that extends the source's metaphor into absurdity. Clay → Porcelain works because both are kiln-fired and one is the receptacle of the other; the name carries the joke AND a real proposal in the same word. Use "let's call it X" as the inline framing. The name is half the thesis. 4. **Asterisked-footnote-as-closer.** Rhetorical question + footnote that calls back to the visual or the absurd register. "Would you use this tool?* / *toilet space ships not included." The footnote does the deflection that lets the rhetorical question land warmly instead of corporate. Without the footnote the question reads as a content-marketing CTA; with it, the question reads as an invitation. **Tone rules specific to this genre:** - **Confident absurdity around a real thesis.** Whimsical packaging, serious argument. The contrast sharpens the argument by making readers re-read the serious part. Don't water the thesis down to match the silliness — sharpen it so the silliness has something to wrap. - **One thematic emoji, doing pun work.** 🚽 earns its place because it IS the Porcelain pun confirmed. Generic emojis (🚀💡✨) are out in this register. - **Selective ALL CAPS once.** "a LOT" — single deployment, doing emphasis work without markdown. **Anti-patterns specific to this genre:** - ❌ Whimsical packaging without a real thesis underneath. The absurdity needs an argument to wrap; otherwise it's just a bit. The Seussian register is a delivery vehicle, not the payload. - ❌ Pun-chain that doesn't extend the source's material logic. Random punny names ("Schmlay," "ClayBot 9000") miss the move — they punch the source instead of extending it. Clay → Porcelain works because it follows the kiln-fired-ceramic chain into the receptacle that receives what Clay produced. - ❌ Footnote that explains the joke instead of extending the absurd register. Annotate-the-bit footnotes kill the bit. The footnote should be funnier than the question, or at least more specific. - ❌ Stacking multiple cross-domain analogies. One does the proof. Two dilutes — the reader starts grading them against each other instead of agreeing with the argument. ### Long-form LinkedIn: interview field notes The field-notes-from-a-conversation format (Sally, Christophe, Jayesh, Alex, Comstock, etc.) is its own genre. It's how Zig turns every in-person encounter at a conference or show into a standalone teaching artifact. Recognizable anatomy: 1. **Time marker + scene** — "Last week at #Dreamforce, I got 30 minutes with…" 2. **Compressed credential line** — who they are, one sentence. Resist the urge to pile on a backstory paragraph; cut it even when it's interesting (the Jayesh OpenAI-when-they-were-9-people story got axed for exactly this reason) 3. **"Here are my #[Event] field notes from our chat:"** — the bridge into bullets 4. **3-5 emoji-bulleted field notes** — each is [short rule] + [1-2 paragraphs with concrete details and a direct subject quote where possible] 5. **Subject-quote outro** — let the interviewee have the last word, unadorned, right before credits. "Jayesh said it best: '.'" 6. **Merged credits paragraph** — photographer + connector + the person who brokered the intro, in one line **Ruthless compression.** Cut anecdotes and backstories that don't earn their line count. A post that feels complete at 400 words will almost always read stronger at 320. When in doubt, cut the peripheral joke (the convention robot, the side anecdote) before you cut the teaching. **Network cross-references are load-bearing.** Tag the panels u hosted, the podcast guests u've referenced, the connectors who set up the meeting. These posts aren't just writeups — they're networking artifacts that signal the density of Zig's professional graph. Hayley-style credit to the person who brokered the intro is non-negotiable. **Thematic emojis, not literal ones.** ♟️ (chess for strategy) beats ⚙️ (gear for mechanism). 🏛️ for institutional power beats 🏢 for office. 🚜 for slow- steady scaling beats 📈 for growth. Test: could this emoji move to a different bullet and still make sense? If yes, it's too generic — pick a more specific one. **Emoji budget for this format.** 3-5 for the bullets, 1-2 for the scene or closer, occasional 🐸✨ identity stamps. Total often runs 6-9 per post, which overflows the short-post budget above. That's fine for this genre — the density is what makes the field notes scannable. **Rhetorical first-person questions mid-post** for peer framing: "How can I quickly, safely, and accurately get the right signals back to my agent?" Signals "I'm wrestling with this too" rather than "here's the answer." **Crystallizing flip lines** are where the teaching lands. One-sentence reframes inside a bullet that make the abstract concrete. "You don't work on Salesforce, Salesforce works inside of your pre-existing infra." Plant at least one per post. Two is better. **Accessible framing over insider framing.** "Demo isn't production" > "Day 1 ≠ Day 2." When the choice is between a technically precise phrase and a universally readable one, prefer the universal one. ### Long-form LinkedIn: narrative-story variant A variant of the interview field-notes format where the 3-5 emoji-bulleted rules get replaced by 1-2 hyperspecific anecdotes that CARRY the thesis (not decorate it). Use when the interview's best material is a story — bobcat in a patient's kitchen, engineer raising a hand w/ "uh, we have MCPs for those" — rather than a rule-set. Exemplars: Alex Waddell TDX post (2026-04-22), Andrew Comstock TDX post (2026-04-23). **Anatomy, diverging from pure field-notes:** - Replace the 3-5 emoji-bulleted rules with 1-2 anecdotes that do the argumentative work. The bobcat story isn't color; it's the reason the "compounding return u can't buy by cobbling together 15 vendors" thesis earns its space. - Selective ALL CAPS on 2-3 words per post (verbs / prepositions that reframe specificity). "IN the home," "INTO those houses," "SINGLE Salesforce org," "NO production path," "the EXACT data." - Visual-comedic devices — a quoted mock form field ("☑️ BOBCAT IN HOUSE"), an ALL-CAPS absurd utterance, a fake diagram line. Turns the anecdote into a visual scene. - Callback-as-unit-of-measurement — coin a phrase mid-post that reuses an earlier detail as an adjective ("bobcat-in-house level chaos"). Rewards re-reads. Signature Zig move. - State the business thesis plainly once the story has earned it. Positive-framing over avoidance-framing. - Strategic pivot to the larger narrative (optional) — 1-2 sentences max. - Credits microformat — photographer routine line ("Adam Noble on selfie duty as always") + broker thank-you (honor what they brought to the chat, not just access) + on-site coordinator with pronouns where given ("Vera Wang (she/her)"). Emoji grammar: 📸 to lead, ✌ 🤙 for warm register, or drop emojis entirely on reflective posts. **Length:** 250-350 words. Longer than pure field-notes (120-180), shorter than Dev Interrupted long-form articles. **Emoji budget is post-register-dependent.** Wild / comedic anecdote posts run 6-8 emojis (Alex final: 🐆 ☑️ 😉 📸 ✌ 🤙). Earnest / reflective posts run 1-2 (Comstock final: just 📸). Same format, different density. Read the piece's register and calibrate. **"u" vs "you" is also register-dependent.** Scrappy / advice clauses take "u" ("stop trying to be full-stack on every problem"). Reflective / earnest posts stay consistent with "you" throughout. Mixing mid-post reads unsteady unless deliberate. **Also at home in this format:** - Linking out to adjacent thinkers — cite a related writer's work with a URL and integrate their vocabulary into your framing. Example: Comstock final links to Alberto Romero García's "The Shape of Intelligence" and uses his "star-shaped intelligence" phrase to reframe Comstock's skill-floor point. Shows synthesis, not name-dropping. - Grade typography: "A-" / "C+" / "B-" over "A-minus" / "C-plus" / "B-minus." - Earnest phrasing for quote intros: "a line I really resonated with" over "a line I can't stop repeating." ### Builder devlogs (messy-in-the-open) A third long-form LinkedIn genre, distinct from short-form hooks and interview field notes. Use when sharing something you found, tinkered with, and want to kit-up for other builders. ~200-300 words. Scrappy, not arced. **Anatomy:** 1. **Contradiction or scene opener** (1-2 sentences). "Most prompt tricks don't stick in my kit, but I'm having a lot of luck with this one." In-progress warmth beats completed declarative when the practice is ongoing. 2. **Lead with the idea** (not your use of it). Name the concept, cite the researchers / source with a credential marker ("as reported in Sakana AI by researchers Kou Misaki and Takuya Akiba"). Give readers a one-line accessible tagline for the idea. 3. **Claim / fix**, distilled to two or three sentences. "The claim: X. The fix: Y." 4. **What I built / tinkered with.** The deterministic thing you made out of it. Concrete protocol details (commands, axes, math). Close the paragraph with a positive-framing payoff line ("It's on-demand X, Y, Z -- all in one"), not an avoid-framing one. 5. **Resource stack.** 2-4 links on consecutive lines. Action-forward labels ("Grab the skill" > "Writeup link"). Thematic emojis where they add meaning (see below). Pre-sell what's at the link, don't summarize it in the body. 6. **Optional meta-flip closer.** If you used /ssot to write the post itself, share the forced choices. "For example: I used /ssot to write this post. Forced flips include: opening=contradiction, voice=new-tool-in-kit..." Signals "try this yourself" without saying it. Use "For example:" as the inline self-reference tag, not "Meta:" (which labels it as self-conscious). 7. **CTA that names a specific reader need.** "Building with LLMs daily and looking for easy ways to inject true randomness" beats generic "building with LLMs daily." Pre-qualify the audience you want. **Link styling:** - Action-forward verbs: "Grab the skill" / "Steal my config" / "Clone and ship" -- not "Link" or "Writeup" - Thematic emojis layered for observant readers: - 🧰 for tool / skill to grab - 🎏 for Sakana AI content (fish thematic nod, sakana = 魚 in Japanese) - 🔗 for neutral / paper / generic source - Stack links on consecutive lines, not inline. The resource block is a distinct unit readers can copy wholesale. **Drop the time marker** ("Last week...", "Yesterday...") when the piece is about an ongoing practice rather than a one-time event. The post should sound like something you're still using, not something you finished. **Genre anti-patterns** (all banned in this format, some also banned elsewhere): - "Here are the 5 lessons I learned from X" arcs - Over-confessional arcs centering self-discovery - Gotcha-ism AI storytelling - Summarizing the content of your own links - Apology-register openers - Forced arc/climax structure ### Carousel / social promotion - Lead with teaching, not just a headline - Each slide should standalone but create swipe momentum - Minimal text per slide — 10-20 words main point - Close with specific CTA (link, comment prompt) --- ## Visual style references Andrew's posts increasingly carry a recognizable visual register alongside the prose. The canonical mood-board lives at [reference/](reference/) in this skill folder. Anyone generating an image to accompany an Andrew post pulls every file in that folder as image-to-image style references so the look stays cohesive across the feed. **The aesthetic in two sentences**: saturated cobalt-marigold-orange palette, 3D clay / matte / textured surfaces (knit, corduroy, beaded), centered subjects on flat split backgrounds. Retrofuturist 60s-70s pop-art register, design-forward, slightly surreal. **Convention for adding new generations** to the reference set: when an image lands well (post ships, aesthetic holds), copy the final file into `reference/` with the name pattern `YYYY-MM-DD-.png`. The next generation will pull it as a style anchor automatically. See [`reference/README.md`](reference/README.md) for the full convention, the no-go list (one-off brand fusions, rejected drafts), and the catalog of current files. **For implementation** — see [`/openrouter`](../openrouter/SKILL.md) → "Image generation" for the canonical multi-reference image-to-image call shape. nano-banana 2 (Gemini 3.1 Flash Image) handles the multi-reference set well; the more diverse the prior set, the more the generation captures the register without copying any single source. --- ## Session Receipts Dated log of writing-craft sessions that contributed rules and examples to this skill. When you're not sure where a rule came from, check here. - **2026-04 TDX session.** Dev Interrupted long-form article on Salesforce TDX 2026 (v3 -> v7 iterations). Accompanying LinkedIn narrative-story posts for Jayesh Govindarajan, Alexander Waddell, Andrew Comstock. LinearB brand newsletter-promo post. Build vs. Buy event-swap graft into the article. Separate SSoT-skill devlog post. Voice calibration arc: v4 teacher-frame collapse -> v5 conspiratorial-whisper fix -> v6 headless+loop title (rejected as grim) -> v7 practitioner POV locked. Published Alex and Comstock posts became exemplars for the narrative-story LinkedIn variant. Moves added to SKILL this session: compression-across-surfaces, cross-reference-with-vocab-import, emoji-budget-by-register, earnest-over-performative phrasing, grade-typography, broker-credit-nuance, selective-ALL-CAPS-emphasis, narrative-story field-notes variant, apology-register-ban, teacher-metaphor-one-flash-max, gotcha-ism-AI-storytelling-ban, multi-surface compression rule. - **2026-04-27 Porcelain post.** AZ solo-drafted a short LinkedIn post on Clay/sender-receiver email asymmetry, with a Nano Banana toilet space-ships image companion. Asana task gid 1214308984308987. Brought to zig-agent-copy as a "learn from this" exemplar — not for rewrite. Reading the post locked in the **short-form whimsical thesis post** as a named genre (distinct from short-form / field-notes / narrative-story / builder-devlog), and named four moves the skill hadn't called out yet: pun-chain-naming-gambit (Clay → Porcelain), cross-domain-analogy-as-the-proof (cybersecurity attacker/defender parallel), confident-absurdity-around-a-real-thesis (Seussian packaging that sharpens the argument rather than diluting it), asterisked-footnote-as-closer (rhetorical question + footnote callback that subverts the corporate CTA register). Whimsical / Seussian register became its own format-specific section under Format-Specific Notes. ## The Zig Test Read the content back. Does it sound like a very online, very smart, very enthusiastic person who genuinely cannot stop talking about agentic engineering and wants you to care too? If it sounds like it could come from any brand account, rewrite it. Does it show the messy path to understanding, own its mistakes, and invite others into the discovery? If it just presents conclusions, rewrite it. Could you imagine Zig saying this out loud while pacing around a conference floor in a cow sweater? If not, rewrite it.