
writing-fragments
Interviews the user to mine raw, unstructured writing fragments and appends them to a markdown file for later shaping.
Lift-25ppscreen tier · directional
88
Trustsecurity · quality · license · upkeep
Scan Saferisk 0/100
LicenseMITserved via MCP
Skill-matched tasks (generated, public + held-out) · glm-5.3-flash@cf
-25pp95% CI [-63, 0]unverified · screen tier
trigger rate 100% · 24 trials
A/B outputs — median trial per arm
Same task, same model: without the skill (left) vs with it (right).
writing-fragments--m1
baseline · 1.00
These fragments are already doing a lot of quiet work — they're specific, sensory, and emotionally oblique, which is exactly what makes a personal essay sing rather than tell. Here are the threads I see worth pulling: **1. Learning a place through its logistics** The bus schedule before anyone's name is your strongest line, I think. It captures something true about new-city life: you master the *systems* of a place before its *people*. There's an essay in that — how intimacy with infrastructure substitutes, temporarily, for intimacy with humans. You could trace it: the bus route, the grocery store layout, the trash pickup day — the way competence in a place arrives before belonging does. **2. Objects as identity** The red fob "felt like a passport" is a gorgeous image — a small object conferring citizenship you don't fully feel yet. What else carried that weight? The key itself, a transit card, a local ID? You could build a thread around the props of belonging: things that say *you live here* before you believe it. **3. Being a spectator before being a participant** Watching the family argue in another language while eating alone at a counter — that's the posture of the newcomer: audience, not actor. And the park with wet concrete and cut grass is you *receiving* the city rather than shaping it. There's a thread about the shift from observer to participant — when did you first do something *in* the city rather than just watch it? **4. Performing adjustment** "Someone said I seemed lost, and I laughed too long" — this is the emotional center, honestly. The laugh is the essay's wound: the gap between how you appear and how you feel, and the small performances of being fine. "Laughed too long" is doing so much — it implies you heard the truth in their observation. **5. Language and translation** The family arguing in a language you don't speak, the bus schedule as a text to memorize, the coworker's comment you had to translate into a joke — there's a thread about all the ways you were illiterate in this new place, and how you slowly learned to read it. **A possible shape:** You could structure the essay around *firsts* — first meal alone, first bus ride, first conversation — with the "laughed too long" moment as the turn, and some later moment (unwritten yet?) where the city stopped feeling foreign as the resolution. A question worth sitting with: what's the moment you *don't* have yet — the one where you realized you'd arrived, or realized you never fully would? That missing fragment might tell you what the essay is really about.
with skill · 1.00
These are strong fragments already — there's real material here. Before we dig in: where should I save the fragment file? A path like `moving-city-fragments.md` works, or tell me wherever you keep your notes. Meanwhile, here's what I see pulling on these threads: **1. Translation / being outside language.** The family arguing in another language, laughing too long at "you seem lost" — there's a thread about being a non-participant in other people's fluency. You can hear everything and understand nothing. Worth mining: what else did you overhear that first month that you couldn't parse? **2. Objects as identity substitutes.** The red fob "felt like a passport" is your most loaded image — a key that certifies you belong somewhere you don't yet. What other objects carried weight? A transit card, a grocery receipt, a name tag at work? There may be a whole cluster of "things that stood in for belonging." **3. Systems before people.** "I learned the bus schedule by heart before I knew anyone's name" is almost a thesis statement in disguise. Memorizing routes, store hours, trash day — mastering the *infrastructure* of a place as a way to feel competent while the human layer stays empty. Did you deliberately learn anything else by heart? **4. The smell as anchor.** "Wet concrete and cut grass" — sensory details are what make a new city specific rather than generic. What did it smell like at night? What did the first place that felt familiar smell like? **5. Performing okay-ness.** Laughing too long is the sharpest fragment of the five — it's a confession disguised as an anecdote. The gap between how you appeared (fine, amused, adapting) and how you felt (lost, seen, exposed). Was there a moment the performance cracked? **A possible leading word:** something like *untranslated* — the state of living somewhere you can't yet read. Or *counter-height* — the literal position of eating alone, eye-level with strangers you don't talk to. If one of those clicks, it could name the whole piece. Which of these feels closest to what you actually want to write about? And tell me where to save the file so I can start capturing as we go.
writing-fragments--m2
baseline · 1.00
Here are some raw threads and directions from your fragments: **Threads within the notes** - **Scarcity & demand** — heirloom tomatoes gone by 9am says something about what local shoppers want and how small the supply is. Could become a piece on "get there early" culture or on why local growers can't scale. - **Economics of small food businesses** — the permit fee complaint is a policy/cost story. Food trucks are a good lens for how fees, rents, and regulation squeeze margins. - **Closed loops / sustainability** — the compost-to-farm-to-restaurant cycle and the rainwater garden donating surplus are both circular-economy stories. These two could pair naturally. - **Supply chain shifts** — the bakery switching to ancient grains from a regional mill is about provenance and taste, but also about rebuilding regional grain infrastructure. **Possible angles for the newsletter** 1. **"The loop" feature** — follow one ingredient or one waste stream around town (restaurant → farm → restaurant; garden → shelter). Concrete, visual, feel-good. 2. **"What it costs to cook here"** — a series on the unglamorous economics: permit fees, mill prices, market stall costs. More investigative, could anger some subjects but builds credibility. 3. **Timing piece** — "9am is too late": a practical guide to the farmers market built around what sells out and when, with vendor interviews. 4. **Provenance trend piece** — ancient grains, heirloom varieties, rainwater: is the whole scene drifting toward heritage and sustainability, or are these outliers? 5. **One-block portrait** — pick a single street or market and use all five fragments as texture in one day-in-the-life piece. **Gaps to fill before choosing** - Who are these people by name? Named sources make any of these stronger. - Is the permit fee complaint widespread or one person's gripe? Check with other trucks. - Does the shelter actually want the produce, and how much is donated? If I had to pick: the closed-loop story (#1) is the most distinctive and uses two of your fragments; the economics series (#2) has the most legs as a recurring feature.
with skill · 0.00
writing-fragments--m3
baseline · 1.00
Here are some threads worth pulling on — not an outline, just raw material. ## Tensions **Sleep as janitor vs. sleep as artist.** Your notes contain two competing images: sleep as waste removal (mechanical, housekeeping) and sleep as memory consolidation (creative, selective). The brain both *cleans* and *curates* — it flushes debris while deciding which memories deserve permanence. That's a rich contradiction: the most passive-seeming state we enter is also where our identity gets edited. **The nap paradox.** "Some studies show benefits, others show none" is a gift for a piece about sleep — because it mirrors how people actually experience naps. Sometimes you wake restored; sometimes you wake groggier than before. The science's inconsistency validates lived experience. You could argue naps are a Rorschach test: we read into them what we need. **Light as saboteur.** Blue light delaying melatonin is small on its own, but it becomes interesting as a *modern intrusion* — the tension between an ancient biological rhythm and a screen held six inches from your face at midnight. Sleep as something we're actively talking ourselves out of. ## Metaphors worth testing - **The night shift.** While "you" are offline, a crew works: flushing waste, filing memories, running threat simulations. Consciousness as the day manager who never sees what happens after closing. - **Sleep as editor.** Consolidation isn't recording — it's cutting. Deep sleep decides what stays in the final draft of you. - **Dreams as rehearsal space.** The threat-simulation idea suggests dreams are a flight simulator: cheap crashes, no real stakes. (Counterpoint worth exploring: why do so many dreams rehearse embarrassments rather than predators?) - **Deprivation as slow leak.** Emotional dysregulation from lost sleep suggests sleep isn't restorative *additionally* — it's load-bearing. Remove it and the structure sags in specific, predictable places. ## Entry points 1. **Start inside the janitor's shift.** Open mid-sleep, from the brain's perspective — the glymphatic system doing its rounds — then zoom out to the person who will wake up slightly rewritten. 2. **Start with the failed nap.** A personal or observed scene: waking from a "restorative" nap feeling worse, then discovering the science can't even agree. An honest piece about sleep could begin with its messiest data. 3. **Start with the 2 a.m. screen.** The blue light note is the most culturally resonant — everyone recognizes the guilt of it. Use it as a doorway into the deeper machinery. 4. **Start with what sleep takes.** We frame sleep as gain (rest, repair), but consolidation is also subtraction — forgetting is a feature. A piece about what the brain *discards* overnight might be fresher than one about what it preserves. ## A possible through-line (not an outline, just a question) What if the piece's spine is: **sleep is not the absence of thought but a different kind of thinking** — one that cleans, edits, rehearses, and regulates, all without a witness? Every one of your six notes fits under that single reframe. Which of these pulls at you? I can go deeper on any thread.
with skill · 1.00
Before we dive in — one housekeeping question: where should I save the fragments file? (A path like `sleep-memory.md` is fine.) I'll capture everything you say into it as we go.
Now, your notes. Let me push on them, because there's more tension in there than you might think:
**The central tension I see:** your fragments split into two stories that fight each other. "Sleep clears metabolic waste" and "memories consolidate during deep sleep" are the *tidy* story — sleep as maintenance crew, night shift janitors hosing down the brain. But "napping studies are inconsistent" and "sleep deprivation impairs emotional regulation" are the *messy* story — sleep as something we barely understand, where the same nap helps one person and does nothing for another. Which piece is this: the tidy story or the honest one?
**Metaphors worth mining:**
- *The night shift* — the brain's janitorial crew, glymphatic system as plumbing. Compelling but overused; could be subverted ("the janitors also rewrite the ledgers while they clean").
- *Sleep as editor* — it doesn't just store the day, it cuts it. Consolidation is deletion as much as preservation. What gets kept? What's the brain's editorial policy?
- *Melatonin as the body's dusk signal* — blue light as a lie told to the pineal gland, midnight at noon. There's a piece about modern light as time-travel fraud in there.
- *Dreams as flight simulator* — threat-rehearsal dreams. The brain running disaster drills while the body is parked.
**Entry points / hooks:**
- Start with the inconsistency, not the consensus. "Some naps help. Some do nothing. Nobody knows why" is a more honest opening than another glymphatics explainer.
- Start with the emotional one: sleep deprivation impairs emotional regulation — meaning a bad week of sleep makes you a slightly different person. Who were you before the bad week?
- The janitor-and-forger paradox: the same organ that cleans the brain also edits its memories. Cleaning and rewriting in one shift.
**Leading word candidates** (a term the whole piece could hang on):
- *The night shift*
- *Overnight edit*
- *Consolidation debt*
- *The dusk signal*
Which of these tugs at you? And tell me: is this piece for you personally (do you have a bad-sleep story of your own?), or is it explanatory? That changes which fragments survive.writing-fragments--m4
baseline · 1.00
Here's a read on your raw material before you structure anything: ## Provocative throughlines **1. The "luxury good" framing is your sharpest blade.** The economist quote + the 60%-would-take-a-pay-cut stat pair beautifully: remote work behaves exactly like a luxury good — people pay a premium for it. That reframes the whole debate. RTO mandates aren't about productivity; they're about who gets to keep an amenity that has real market value. An op-ed could argue RTO is effectively a pay cut imposed on workers — the mirror image of the pay cut they'd voluntarily accept. **2. The vacancy number is a fiscal story, not a real estate story.** 22% downtown vacancy isn't interesting as empty offices — it's interesting because mayors are openly admitting transit budgets depend on commuters. That's a provocative admission: cities built a funding model on forced presence, and now they're lobbying to restore the presence rather than fix the model. Follow that thread: who actually bears the cost of remote work — workers, landlords, or transit riders? **3. The 20%→15% drop is doing quiet work.** It undercuts both camps. It's not the "remote work is dead" narrative (15% is still enormous, historically), but it's also not "remote work won." The honest throughline: this is a negotiation, not a revolution — and the trend line is a bargaining outcome, not a technological destiny. ## Ass worth interrogating (assumptions) - **That downtown vacancy = loss.** The suburban co-working fragment hints at the counter-assumption: maybe activity is dispersing, not disappearing. But "still niche" is the honest qualifier — don't let the op-ed overclaim a suburban boom the data doesn't support. - **That the 60% survey stat means what it seems to.** Stated willingness to take a pay cut is cheap; revealed behavior is different. Also: *which* remote workers? If it's high earners (per the economist), the stat may describe a privileged cohort, not workers broadly. - **That mayors' RTO pushes are about transit at all.** It could be cover for downtown commercial tax bases. Worth asking what's really being funded. ## Gaps in your material - **No worker-side economics**: who is remote? Class, industry, geography. Without this, the "luxury good" line can't land with evidence. - **No counterfactual**: what would downtowns look like if remote work fully reversed? Is 22% a crisis or a correction from an artificially tight pre-2020 market? - **No outcomes data**: nothing on productivity, retention, or what companies actually get from RTO. Right now every fragment is about *presence*, none about *performance*. - **The suburbs fragment is orphaned.** It's your only forward-looking piece; either build the "cities follow people" argument around it or cut it. The strongest op-ed thesis hiding in this pile: *remote work didn't break cities — it exposed that cities were financed by a commute, and now everyone's fighting over who pays for the exposure.*
with skill · 1.00
Before we go further — where should I save the fragments file? (e.g., `remote-work-cities.md`) I'll capture everything from this first message there. Meanwhile, here's the grilling on your raw material: **Most provocative throughlines** 1. **The transit hostage crisis.** Mayors pushing RTO *to fund transit* is the spiciest fact you have — it inverts the usual framing. Cities aren't asking people back because downtowns are dying; they're asking commuters back because the farebox is. That's a throughline: remote workers are being conscripted into municipal finance. The op-ed could hang on that — the commute as a hidden tax. 2. **Remote work as luxury good.** The economist's line + the 60%-would-take-a-pay-cut stat pair beautifully: people are paying for it *twice* — in salary and, via their taxes/commutes, in subsidizing the city they no longer use. Is remote work the first major workplace benefit that workers buy back from their own employers *and* their own city? 3. **The vacancy is the story, not the worker.** 22% office vacancy vs. only a 5-point drop in postings suggests the market already repriced — the fight over RTO is happening *after* the money moved. Downtowns aren't waiting for a policy debate; they're already something else. **Assumptions worth interrogating** - That downtown vacancy = urban decline. Suburban co-working growth hints work is *dispersing*, not disappearing. Cities may be losing their monopoly on proximity, not on employment. - That the 60% survey stat means what it seems to. Would-take-a-pay-cut is stated preference; the postings data is revealed preference — and revealed preference says employers are pulling back. Who's right? - That mayors' RTO pushes are about transit at all — could be real estate exposure (cities often hold bonds backed by office values). Worth naming out loud. **Gaps** - No human. Every fragment is a number or an official. Who is the vignette? A suburban co-working regular, a transit driver, a downtown small-business owner? - No counterfactual: what *should* a city do if commuting is permanently down 30%? The op-ed provokes better if it answers rather than laments. - The "luxury good" claim needs a mechanism — luxury for whom, and what happens to the people (service workers, transit riders) whose livelihoods rode on the commute? **Leading-word candidates** (the coinage that could carry the piece): *the commute tax*, *conscription commuting*, *the farebox hostage*, *proximony* (proximity + alimony — the payments cities demand for a relationship that's over). Which of these makes you itch? And where do I save the file?
writing-fragments--m5
baseline · 0.85
These are good fragments — and what's interesting is that the surface subject (cooking) is barely the subject at all. Let me point at the live wires I see: **The mother line is your strongest vein.** "Mom never let me in kitchen, said I'd slow her down" isn't a cooking memory — it's about being excluded from inheritance, from a domestic lineage most people absorb by osmosis. That's why learning to cook as an adult feels different from learning anything else as an adult: you're doing it without the person who was supposed to teach you. And it quietly recontextualizes "burned rice first time, ate it anyway" — you ate it anyway because nobody was going to feed you. That note is more defiant than it looks. **The ex's grandmother is the strange one.** You inherited a recipe from a lineage you were never part of, and you *still make it*. That's a small mystery worth sitting with: why keep a connection to a family that ended? Is it the food, or is it proof that someone's grandmother once handed you something? It might be the only "passed down" thing you have, which makes it poignant next to the mother note. **The loneliness is doing more work than the skill.** "Silence of cooking alone after work" and "grocery store at 9pm, buying one onion" are the same image from two angles — adult solitude rendered in miniature. One onion for one person is a very specific, very legible loneliness. I'd resist resolving this into "and then I found joy in cooking!" The aloneness might be the honest core, and the skill is just what happens inside it. **The friends-impressed-by-scrambled-eggs note has a hidden barb.** Being praised for something basic can feel good *and* slightly humiliating. Worth asking: did you let them believe you were more capable than you were? Is there a small performance in your cooking that the silence at home doesn't have? **The tattooed YouTube chef** is your only note about learning from strangers rather than intimates — the modern condition. A spoon tattoo is a nice concrete detail: someone so devoted to this that it's on their body, teaching you through a screen. There's a contrast there: devotion you can witness but never receive. **"Knife hand is clumsy but getting steadier"** is the only note about the body, and it's the most forward-looking one. Everything else looks back (mom, ex, burned first attempt). This is the one sentence that's about becoming. If I had to name the piece hiding in these: it's not about learning to cook. It's about an adult teaching themselves a form of care they were never shown — and the question underneath everything is whether cooking for and by yourself counts, or whether it only becomes real when someone else tastes it. The friends note and the silence note are arguing about exactly that. What's your gut reaction — does the mother angle feel like the piece, or too heavy?
with skill · 1.00
Happy to dig in. One quick thing: where should I save the fragment file? (e.g. `cooking.md`) — I'll append everything there as we go, and you can edit it between turns. Meanwhile, let's start grilling. Your notes are good raw material, and a few of them are already doing more work than they look: **The one that's secretly the spine, I think:** "mom never let me in kitchen, said I'd slow her down." That's not a cooking memory, that's an origin story. Everything else — burned rice eaten anyway, clumsy knife hand getting steadier, cooking alone in silence — reads differently once that line is on the table. You're not learning to cook; you're learning to occupy a room you were explicitly kept out of. Does that ring true, or am I over-reading one line? **The one with the most tension:** "recipe from ex's grandmother, still make it." There's a whole essay hiding in that. Why do you still make it? Sentiment, or is it just a good recipe? Does making it feel like keeping something or like theft? What's the dish? **The one that's a scene, not a note:** "grocery store at 9pm, buying one onion." That's a complete image of adult solitude. Tell me more about that trip — was it the first time you cooked for yourself? What did the single onion mean that night? **Questions to loosen things up:** 1. The YouTube chef with the spoon tattoo — what did he actually teach you, beyond technique? Was he the first person who made cooking seem *permissible* to you? 2. "Friends impressed by scrambled eggs" — impressed, or were you embarrassed that the bar was that low? Those are different feelings. 3. When you burned the rice and ate it anyway — was that defiance, poverty, or not knowing you were allowed to throw it out? 4. "Silence of cooking alone after work" — is that silence peaceful or lonely? Be honest, because the piece changes depending on the answer. And a leading-word hunt: we're looking for one term that names the whole pattern. Candidates so far, half-formed: *the slow-down* (your mother's word turned into your inheritance), *one-onion errands*, *eating the evidence*. None of these are right yet — what's a word or phrase that keeps recurring in how you *think* about this, even privately?
Limitations
- Agent noise: single-model results vary run to run; the confidence interval reflects task-level variance.
- Screen tier (1 trial/arm) is a directional gate, not a verified claim.
- Results are for the SkillzBench Reference Agent on glm-5.3-flash@cf and do not necessarily transfer to other agents or models.
- Task coverage: Skill-matched tasks (generated, public + held-out) only; a skill can shine elsewhere.