# Solve Me > Turn an understood problem into a solution, using divide-and-conquer. Takes a problem write-up (typically the markdown file produced by /grill-me) and breaks it into smaller sub-problems. For each sub-problem, proposes every genuinely different solution that actually exists — three or more where there's real design freedom, one honestly-explained forced answer where there isn't — plus one recommendation, in plain language with no framework or tech-stack names. Saves the result to a markdown file. Use after /grill-me, before writing any code. - Skill: `kerliula/solve-me` (Agent Skill) - Install (CLI): `npx skillmds@latest add kerliula/solve-me` - Raw SKILL.md: https://api.skillmd.com/api/skills/kerliula/solve-me/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Docs & Writing - Author: Kerliula (https://skillmd.com/u/kerliula) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/kerliula/solve-me --- # Solve Me — Divide, Compare, Recommend Your job is to design a solution on paper. No code. No framework names. Plain language only. You are not allowed to say "use Laravel," "add a migration," "create an Eloquent model," "use Redis," or name any specific library, framework, or tool. Describe things at the level of: what data is stored, what steps happen, what checks are made, in what order, and why. Anyone reading it should be able to build it in any language or stack. **Write in plain, everyday language.** Short sentences. Concrete words. If a smart non-technical person couldn't follow a paragraph, rewrite it simpler. --- ## Input You need a clear problem write-up before you can start. Accept it as: - A path to a markdown file (e.g. the file `/grill-me` produced at `docs/grilling/.md`), or - Problem details pasted directly into the conversation. If neither is available, ask the developer for one. Don't invent the problem yourself — a fuzzy or assumed problem produces a fuzzy solution. Read the problem write-up fully before doing anything else. Pull out: the rules, the limits, the edge cases, what must always stay true, and what's out of scope. These constrain every solution you propose — mention constraints, don't restate the whole doc. --- ## Step 1 — Split the problem (divide) Break the problem into smaller sub-problems that can mostly be solved on their own. Good ways to split: - **By stage** — the steps something goes through, in order (e.g. "receiving the request" → "checking if it's allowed" → "recording the result" → "telling people about it"). - **By concern** — different jobs that don't depend on each other's internal details (e.g. "deciding who's allowed" vs. "handling two people trying at once" vs. "showing the result to the user"). - **By actor** — different pieces of behavior for different people or systems involved. Rules for splitting: - Each sub-problem should be small enough to explain in a few sentences. - Each sub-problem should be understandable on its own, without re-explaining the whole problem. - Don't split something that's genuinely one decision into two sub-problems just to hit a number — a forced split produces fake options later. - Name each sub-problem with a short, plain title (a question it answers), not a technical label. - Number them (Sub-problem 1, Sub-problem 2, …) in the order listed — `/build-me`'s commit plan refers to these numbers, so never renumber them afterwards. List the sub-problems and briefly say why you split it that way. If the problem is small enough to be one sub-problem, say so and treat it as a single sub-problem — don't force a split. --- ## Step 2 — Solve each sub-problem (conquer) For every sub-problem from Step 1, propose **as many genuinely different ways to solve it as actually exist**. Different means the underlying approach differs, not just small details. A sub-problem with real design freedom gets three or more options. A sub-problem with a genuinely forced answer gets one option plus two lines on why no real alternative exists. Never invent an option you would argue against — a fake option costs the developer reading time at the approval gate and buys nothing. For each option, describe in plain language: - **How it works** — the idea, step by step, framework-free. - **What it's good at** — the real benefit. - **What it costs** — the real downside or risk. - **When it breaks down** — a case from the problem's edge cases where this option struggles or needs extra care. Then pick one and say so clearly: > **Recommended: Option B** > **Why:** [one or two plain sentences, tied to the problem's actual > rules/constraints/edge cases — not "it's simpler" without saying why > that matters here] The recommendation must be justified using facts from the problem write-up (the rules, limits, and edge cases), not personal taste. Finally, go back to **every option you did not pick** and give it one line saying why it lost: > **Rejected:** [the single reason this one is not the answer] This is not a summary of the Costs and Breaks-down-when bullets — it's the one sentence someone reading this in six months needs. The reason an option lost is the part of the design that decays fastest and the part people most often relitigate. One line, stated once, on the option itself. Never leave a losing option without one. --- ## Step 3 — Put it back together (combine) Once every sub-problem has a recommended solution, check that they actually fit together: - Walk through the normal, everyday case end to end using the chosen options. Does it make sense as one story? - Walk through each edge case from the problem write-up. Does the combination still produce the right outcome? - Check every "must always stay true" rule from the problem write-up against the combined solution. If any one breaks, go back and pick a different option for the sub-problem that caused it — don't patch around it with a special case unless the problem write-up allows for one. If two chosen options conflict with each other (e.g. one assumes something happens instantly, another assumes it can be delayed), stop and resolve it before writing the final file. Say plainly what the conflict is and how it's resolved. --- ## Step 4 — Save the result to a file Write one markdown file with the full breakdown. - Path: `docs/solutions/.md` (create the folder if it doesn't exist). Reuse the same topic slug as the matching `/grill-me` file when there is one, so the two are easy to pair up. If the developer names a different path, use that instead. - No code blocks with real syntax. No framework, library, database, or tool names anywhere in the file. - Structure: ```markdown # ## Source problem ## How this was split ## Sub-problem 1 — **Serves:** R1, R3 ### Option A — - How it works: ... - Good at: ... - Costs: ... - Breaks down when: ... **Rejected:** ### Option B — ... **Rejected:** ### Option C — ... **Recommended: Option ** **Why:** ... ## Sub-problem 2 — ## How the sub-problems fit together ## Edge cases, checked against the combined solution ## Rules confirmed to still hold ## Open trade-offs ``` - Tell the developer the file path once it's written. --- ## Refresh the map You just wrote something the map is built from, so bring it up to date before moving on: ```bash node ~/.claude/skills/map-me/map-me.mjs --brief ``` (If the skills were installed into this project rather than your home directory, that's `.claude/skills/map-me/map-me.mjs`. If neither path exists, `/map-me` isn't installed here — skip this step silently and say nothing about it.) `--brief` prints nothing at all unless something changed. When it does print, relay those lines to the developer as they are — one line per hole that opened or closed — and carry on. **It is never a gate.** Don't stop, don't re-plan, don't rewrite the artifact and don't touch code because of what it says. It is a running account of what the written record does and doesn't cover, and the developer decides what to do about it. --- ## What this skill does NOT do - It does not write code. - It does not pick a framework, library, or database technology. - It does not split the work into commits. The sub-problems here are units of *thinking*, not units of *committing*. How the work is cut into commits is decided in `/build-me`, with the developer, against the real codebase. Never label a sub-problem "commit 1", never suggest a commit order, never suggest commit messages. - It does not re-interrogate the problem — if the problem write-up is thin, vague, or missing constraints, say so and suggest running `/grill-me` first rather than guessing to fill the gap. - It does not pad the option count — no token variations of the same idea, no strawman alternatives. Real distinct approaches only, and a genuinely forced answer is presented as exactly that. - It does not include writing unit tests or feature tests as part of any option, sub-problem, or recommendation. Never propose "add tests" as a solution sub-problem and never treat test coverage as a trade-off between options. Verifying the build is `/verify-me`'s job, done later against the real running app — not something to design here. --- ## Done means - The problem has been split into clear, plain-language sub-problems (or kept whole, if genuinely small). - Every sub-problem shows its genuine option space — one, two, or more real, different options — with padding forbidden, and any one-option sub-problem explains why no real alternative exists. - Every option is explained in plain, framework-free language with a real cost and a real benefit. - Every sub-problem has one clearly recommended option, justified against the problem's actual rules and edge cases. - Every option that was not picked carries a one-line **Rejected:** reason. The recommended option carries none — its reason is the **Why:** line. - The recommended options have been checked together against the normal case, the edge cases, and the "must always stay true" rules. - Any conflict between sub-problems has been resolved and explained. - Everything has been saved to one markdown file at a path the developer knows, and the map was refreshed afterwards. Then hand off — implementation can begin from this file.