# The seven project-folder prompts Copy the prompt for your current step. Replace brackets and keep the approved files attached or connected. These prompts guide your own AI; this template does not itself run an AI. ## 1. Give your idea a clear starting point I want to create [idea] for [audience]. My first useful result should be [first result]. Help me clarify the idea before we build it. Ask up to three focused questions about the problem, my existing material and my constraints. Use my answers to write a short PROJECT.md: who this is for, what they need, the first result, what is outside the first version, and the main unanswered question. Separate facts I supplied from assumptions. Offer one concrete first deliverable, not a large feature list. Explain how I will know it is useful. Show me the brief for correction. Do not invent customer interviews, market demand, prices or results. If you cannot create a file yet, provide its complete text for me to save. **Finished result:** You have a short brief naming the audience, first deliverable and the most important unknown. ### Expert prompt Challenge this idea against its intended user, the situation that triggers the need and my constraints. Compare at most two possible first deliverables. Recommend the smallest one that would teach us whether the approach is useful. Explain the tradeoff and record the decision only after I approve it. --- ## 2. Let your AI read the right files We are setting up my project workspace. Read PROJECT.md and START-HERE.md from the files I attached or the folder I explicitly connected. List the files you can actually access. Quote one short, project-specific line from PROJECT.md so I can verify you read the correct version. Tell me whether you can edit that folder directly, create downloadable replacement files, or only return text in this chat. If a file is missing, ask me to attach it or connect the folder. Do not claim that seeing a folder path gives you file access. Before changing anything, tell me where the revised files will be saved and how I will keep the main copy current. **Finished result:** Your AI confirms a real detail from your file, and you know how updated files will return to your workspace. ### Expert prompt Verify this workspace setup. Identify the authoritative file location, readable files, writable location and any synchronization limits. Distinguish project knowledge uploads from local file access. Propose a short start-of-session instruction referencing START-HERE.md and verify it by reading the current project goal. Do not print credentials or claim untested access. --- ## 3. Turn your brief into a useful project folder Read my approved brief and the sources I provided. Create or update these files: PROJECT.md: audience, goal, first result, scope and constraints. AUDIENCE.md: the user’s situation, needs, objections and evidence. PLAN.md: the first deliverable and a short sequence for making it. REVIEW.md: unresolved questions and checks before building. TASKS.md: one next action, expected output and completion check. DECISIONS.md: only choices I have actually approved. HANDOFF.md: current state and where the next session should start. Fill what the sources support. Mark assumptions and unknowns visibly; ask the most useful missing question. Use inputs/ for sources, outputs/ for drafts and accepted/ for reviewed work. Add at most two specialist files only if this project needs them, and explain why. Save using our agreed route. Return a concise file list with one meaningful decision or question from each file. Do not report empty templates as a completed plan. **Finished result:** Your core files contain useful project-specific information, and TASKS.md names one concrete next action. ### Expert prompt Inspect the project folder for duplicated instructions, contradictory facts and files with no clear use. Propose the smallest structure that preserves the necessary context. Identify which files each next task needs. Keep the approved scope and source evidence intact, and ask before removing existing material. --- ## 4. Choose what you will build first Read PROJECT.md, AUDIENCE.md and PLAN.md. My current constraints are [time, budget, skills, available tools]. My first deliverable is [deliverable]. Relevant examples are [files or links]. Refine PLAN.md around that one deliverable. Separate Must have now from Later. Define two or three checks I can actually perform: what should I see, understand or be able to do? List required inputs and dependencies, with unknowns kept visible. Ask the most useful unresolved question, then show me the proposed scope. Once I approve it, update PLAN.md, TASKS.md and DECISIONS.md consistently. Do not add a new platform, subscription or feature unless the deliverable requires it. **Finished result:** You have an approved first deliverable, a clear Later list and a small set of checks for the result. ### Expert prompt Review the first-deliverable plan for hidden dependencies, unclear ownership and unnecessary complexity. Identify the critical sequence, expected output at each step and relevant tool or usage costs. Recommend the simplest implementation that still satisfies the approved checks. List what can be manual initially and what justifies automation. --- ## 5. Find the gaps before you build Audit the current project files as the intended user and the relevant specialists. Choose roles that fit this project, such as educator, designer, marketer or developer. Review five areas: clear purpose, audience evidence, useful scope, feasible execution and observable acceptance checks. For each, state the supporting evidence, the gap and its effect on the first deliverable. Use Ready / Needs work / Unknown. Rank the three most important corrections. Propose one small real-world check for the weakest assumption. Label simulated user feedback as a hypothesis, not a customer interview. Save the review in REVIEW.md and propose the next task. If you provide a score, explain its criteria and deductions; do not treat a higher self-score as proof of market demand or launch readiness. **Finished result:** You know what is ready, which gap matters most and what evidence or action will resolve it. ### Expert prompt Create a review table with criterion, evidence, status, effect, next correction and verification. If numerical scoring is requested, allocate 20 points to each of the five dimensions and show the deductions. Do not award unsupported points. Recommend the next small experiment that could disprove the weakest assumption, and a clear condition for moving into the first build. --- ## 6. Make something you can open and try Read PROJECT.md, PLAN.md, TASKS.md and the relevant source files. Build only this approved deliverable: [deliverable]. It must pass these checks: [checks]. Work within [time and paid-usage limit, if applicable]. Confirm which files you read, then create the actual artifact in outputs/ or provide it as a download using our agreed save route. Keep unrelated ideas in Later. Tell me how to open or inspect the result. Run the checks your tools support and describe what actually happened. List checks that still need my input. Help me correct the most important issue first. Once I approve the result, keep its version in accepted/ and update REVIEW.md, TASKS.md and HANDOFF.md. A description or mockup is not evidence that a working feature exists. **Finished result:** You have an artifact you inspected, an accepted version and an honest record of the checks performed. ### Expert prompt Review the produced artifact against the approved acceptance checks. Report the exact version, test steps, observed results and unresolved issues. Distinguish a working integration from a static demonstration. Recommend the smallest correction with the greatest effect. Do not expand scope or deploy publicly as a substitute for completing the agreed test. --- ## 7. Keep the context. Continue tomorrow. Close this work session by updating HANDOFF.md, TASKS.md and any approved decisions. Record the current goal, files changed, latest accepted version, checks actually run, unresolved questions and one next action. Name the minimum files the next session needs. Keep HANDOFF.md short. Distinguish files already saved from replacements I still need to download or upload. Do not treat suggestions as approved decisions or drafts as accepted work. Then prepare a resume instruction in SESSION-START.md that asks the next chat to read those files, confirm the approved state and continue the next agreed task. If the project has outgrown its structure, suggest one useful addition and explain its purpose. **Finished result:** A fresh AI session can find your accepted work and identify the correct next action from the saved files. ### Expert prompt Read the current handoff and the files it names. Summarize the approved goal, accepted version, evidence of the latest check and next task. Flag missing or conflicting context before acting. Propose a short plan for this session using only the necessary files. Keep suggested changes separate from approved decisions and end by updating the handoff again.