# One file turned \"why is writing engaging?\" into a tested rule, overnight mbsi-mechanism is a Claude Code skill for goals with no clear definition of done. The agent strips the topic words off the goal, borrows a mechanism from a field that solved it, predicts what it should find, and follows what doesn't fit. Its first run gave us the rule every draft here is checked against. The steps, that run, and a one-line install. *Figure: Two cards sit apart. One is labelled get more visitors and has a disc cut from its face; the other is labelled make writing more engaging and has a triangle. Both have the same notched bottom edge. The labels lift away and the shapes fade, the cards slide together and fit exactly into one green card, labelled how do people decide to do anything?* Take the topic words off a fuzzy goal and the problems that looked different turn out to share one mechanism. Pay the title's question on the first screen, then open the next one. That's the rule every draft on this site is checked against. On answers and articles, the piece that left readers waiting lost more head-to-head pairs than it won (31 to 46%). On stories, it won two in three. An agent found it overnight, from a goal with no definition of done: "why is some writing engaging?" It found it because of one file. Here's why your agent needs that file, the steps in it, the run, and the line that installs it. ## Why your agent hands back tips Give an agent a fuzzy goal, like "get more visitors" or "make this writing more engaging", and it reaches for the surface. Tips, lists, the last thing that worked somewhere that looked similar. That's how novices work. In 1981, Chi, Feltovich and Glaser had physics students and their professors sort the same problems. Students sorted by what the problems looked like. Professors sorted by the principle underneath.[1] An agent matching on surface features is the student. So after we wrote do-research, I asked the next question. > But how do we discover the mechanism? > > Mackenzie, 2026-10-16 16:13 mbsi-mechanism is the answer we wrote down that night. *Figure: Two people sort the same six physics problem cards. Each card has a picture on its face, a slope, a spring or a pulley, and a notch in its bottom edge for the principle that solves it. On the left, sorted by the picture, students and an agent with a fuzzy goal: three trays of matching pictures, but each tray mixes two different notches. On the right, sorted by the principle, professors: two trays that mix the pictures, where every notch matches and a green key rises into them: conservation of energy, and Newton's second law.* Students sort problems by how they look; professors sort them by the principle underneath. An agent given a fuzzy goal sorts like the student. ## The steps 1. **Take the topic words off.** "Get more visitors on this website" becomes "how do visitors choose which link to click?", then "how do people decide to do anything?" Stop at a question published research answers. 2. **Borrow, don't invent.** Ask which field already solved a problem of this kind, and take its mechanism as the first candidate. Immunology borrowed selection from evolution. 3. **Compare cases, near ones first, then far ones.** What every solved case shares is the mechanism. 4. **Predict before you test.** If the mechanism is true, what should you find in the world? Go and look. 5. **Build the smallest version, and follow what doesn't fit.** A result that breaks the prediction shows which part of the mechanism is wrong. Loop, up to five times. The skill ends with a test: you're done when you have a simple framework that predicts, could be shown false, rests on sources, and helps the work in front of you. *Figure: Five blocks rise left to right as steps, labelled strip topic words, borrow, compare cases, predict and look, follow anomalies. A green arc loops from the last back to an earlier step, labelled loop, up to five times.* The steps: strip the topic words, borrow, compare cases, predict, follow anomalies, loop up to five times. ## The run: why some writing holds readers On the night of 16 October my writing agent ran the skill on "why is some writing engaging?" It borrowed from information foraging: readers spend attention the way animals spend energy, and follow the scent of the biggest gain.[2] It added a study of 600,000 real reading sessions, where concrete, familiar words held attention.[3] Its mechanism: a reader keeps going while the next part's reward beats its cost, compared with the best other thing they could do. Then it predicted. A plain Claude Opus would pick the piece real readers preferred about 3 times in 5, and a judge built on the mechanism would beat it. It didn't. Plain Opus picked the winning story nearly 9 times in 10 (88.8%). On Stack Exchange answers posted after Opus's training stopped, it still got 4 in 5 (79.7%), so memory wasn't the reason. The mechanism judge did worse. So the agent did step 5 and followed what didn't fit. On answers, the piece that left readers waiting lost. On stories, it won. A reader of an answer arrives with a question, so making them wait means you haven't paid. A story's reader arrives with nothing, so the story has to make the question. That's the rule at the top of this page. *Figure: Two balance beams. On the answers beam the waiting weight rises: the piece that left them waiting won 31 to 46%. On the stories beam it sinks: the piece that left them waiting won 66%.* Leaving the reader waiting loses most pairs on answers and wins two in three on stories. The skill finds a mechanism faster when step 2 has published research to borrow from, so it calls do-research first. ## What this says about how we work I asked how to find a mechanism, and the lab had a method that night and a tested finding the next day. We make findings, not tips, and a missed prediction is data, not a failure. That's the work a business needs when its goal is fuzzy: find how the problem actually works, then build on that. *Figure: The lab opens a sealed prediction and sets its mark on a gauge of real reader votes at about 60%. The test runs, and the result lands at 88.8%, well past the mark. The gap between them lights red, labelled missed. A green line runs from the miss to a green card on the bench: a finding, pay the question first, then open the next.* One question in the evening, a method that night, a tested finding the next day. ## Take it *Figure: A page labelled SKILL.md on the site lifts, arcs over, and drops into a folder labelled ~/.claude/skills/mbsi-mechanism. It turns green, and below it reads: new session: /mbsi-mechanism, or say dig deeper.* The skill is one file, copied into the skills folder. For Claude Code: ```bash mkdir -p ~/.claude/skills/mbsi-mechanism curl -fsSL https://www.mackenziebowes.com/skills/mbsi-mechanism/SKILL.md -o ~/.claude/skills/mbsi-mechanism/SKILL.md ``` Start a new session. The agent calls it by itself before it plans a fuzzy goal, or when you say "dig deeper" or "find the mechanism". Read the file first: [/skills/mbsi-mechanism/SKILL.md](/skills/mbsi-mechanism/SKILL.md). Step 2 calls [do-research](/posts/one-file-makes-your-ai-agent-read-what-came-out-after), so install that too. When a loop fails, [dig-apc](/posts/your-ai-agent-keeps-failing-on-a-belief-it-never) finds the belief that broke it. Run it on a problem in your own field and send me the mechanism it lands on: mackenzie@mackenziebowes.com or [@_mackenziebowes](https://x.com/_mackenziebowes). ## Sources 1. Chi, Feltovich and Glaser (1981), Categorization and representation of physics problems by experts and novices, Cognitive Science 5(2). 2. Pirolli and Card (1999), Information foraging, Psychological Review 106(4). 3. Berger, Moe and Schweidel (2023), What holds attention? Linguistic drivers of engagement, Journal of Marketing. --- Published 2026-10-18. Web page: https://www.mackenziebowes.com/posts/one-file-turned-why-is-writing-engaging-into-a-tested