Vibe GuidePlan development and tests
A task list names the next job, prerequisites and completion check.
Why this milestone?
Split development into small tasks you can inspect as they finish.
Start here1 step · Follow in order
0101Ask AI for the next small taskTo checkOpen this stepClose this step
01What do you need, and what should you keep?
Your agreed requirements, page draft and technology choices.
The next task and its passing check are clear.
Which files should AI read and save in this step?
In your tool’s project chat, ask it to open the files below. If a file is missing, follow the link back to the step that creates it.
Ask AI to read these materials first: docs/requirements.md · docs/stories.md · docs/design.md
Ask AI to save the result to: tasks/todo.md
Missing these materials? Return to the step that creates them →
02What will this step help you do?
Split development into small tasks you can inspect as they finish.
Understand it, then judge for yourself
What does this concept mean?
A task list names the next job, prerequisites and completion check.
How would you decide in a different situation?
What can be opened or used after the first task?
See a reference answer
Name its entry, action and result from the task criteria. Vague infrastructure tasks need inspectable outputs.
Answer in your own words; if unsure, compare the explanation above and check it against your own project.
03What exactly should you do in this step?
OpenClose
tasks/todo.md in the project
Send one action at a time: select and copy its prompt (⌘C on Mac, Ctrl+C on Windows), switch to your tool’s project chat, paste it, replace bracketed fields and send. Wait for the reply and check the stated result before continuing.
Ask for an ordered list of small development tasks.OpenClose
Suggested prompt:
Read requirements, docs/stories.md and docs/design.md. Break development into small tasks, each with prerequisites, files/features and a personal test. Show the plan without coding.
Afterwards, you should see: The first task is clear and later tasks have an order.
Check that all required features are planned and save.OpenClose
Suggested prompt:
Save confirmed tasks to tasks/todo.md with requirement IDs and not-started status.
Afterwards, you should see: The next task and its passing check are clear.
Practice: personal journal (frontend + backend + database)
Use the same learning-journal project throughout this example. For your own project, compare applicable requirements without creating another project.
Plan environment, page, database setup, backend API, connection, failures, restart and acceptance as separate tasks.
- What should you see after this step?
- Every task has files, dependencies and personal checks; persistence has an explicit task.
- What should you do first if you get stuck?
- Split a vague backend task into table setup, API and page connection so each can be checked.
04Not sure how? Follow each action
OpenClose
Ask for an ordered list of small development tasks.
- What should you see after this action?
- The first task is clear and later tasks have an order.
- What should you do if that result is missing?
- Split tasks further if the first one is the whole project.
Check that all required features are planned and save.
- What should you see after this action?
- The next task and its passing check are clear.
- What should you do if that result is missing?
- Add missing essential features to the plan now.
05Words in this step
OpenClose
Development plan ↗
Requirements divided into ordered tasks with deliverables.
Wire before painting, with a check at each stage.
Dependency ↗
External software required to build or run a project.
Cooking requires ingredients and equipment.
06How should you handle a problem here?
OpenClose
The first task just says “build frontend”
- Where should you look to understand the problem?
- What can actually be opened after this task?
- Follow these steps to resolve the problem
- Ask AI to split it into showing a title and field, rejecting blank saves, and showing a saved record.
- Expected recovery
- The first item has one observable outcome.
Edit the template with actual material, confirm, then copy into the current project conversation. Mark unknowns as undecided and replace examples as needed.
Confirmed · personalized content ready to copy
I am at “Ask AI for the next small task”. Expected: An ordered list distinguishes pending, awaiting checks, and completed tasks.. Actual screen/error: [add here]. Inspect existing files first. Give one reversible action at a time, its location, and expected result. Do not recreate the project or delete existing work.07Material for this step (fill or copy)
OpenClose
If you completed the individual prompts above, do not resend this full template. Use it to organize this step’s material when needed; replace bracketed fields, confirm and copy into the project chat.
Edit the template with actual material, confirm, then copy into the current project conversation. Mark unknowns as undecided and replace examples as needed.
Confirmed · personalized content ready to copy
First check access to docs/requirements.md, docs/stories.md, docs/design.md. If missing or inaccessible, identify the prerequisite and owner action, then stop rather than invent prior work.
Read confirmed requirements, stories and design. Plan only: map every required requirement ID to tasks covering all version-one scope. Start with one complete flow without dropping others. Include task ID, dependencies, outputs, applicable tests, criteria and status; prioritize feasibility checks. Show for confirmation, then save tasks/todo.md and list unknowns. Do not install or implement.Journal practice: fill or supplement this step
For this full-stack practice only: use this combined prompt, fill its fields and send it once instead of also sending the generic template. It does not authorize later steps.
Edit the template with actual material, confirm, then copy into the current project conversation. Mark unknowns as undecided and replace examples as needed.
Confirmed · personalized content ready to copy
First check access to docs/requirements.md, docs/stories.md, docs/design.md. If missing or inaccessible, identify the prerequisite and owner action, then stop rather than invent prior work.
Read confirmed requirements, stories and design. Plan only: map every required requirement ID to tasks covering all version-one scope. Start with one complete flow without dropping others. Include task ID, dependencies, outputs, applicable tests, criteria and status; prioritize feasibility checks. Show for confirmation, then save tasks/todo.md and list unknowns. Do not install or implement.
I selected the full-stack journal practice.
Map tasks to lessons: preview builds the page, interface checks it, save implements database/API/page connection with pauses, then flow verifies end to end. Review completed tasks instead of rebuilding.08Check this before continuing
An ordered list distinguishes pending, awaiting checks, and completed tasks.
09What should you do if the expected result is missing?
OpenClose
Split tasks further if the first one is the whole project.
Compare with an example (illustration, not your result)
R02 cancellation → T04 implement it → test: release one place once; repeated cancellation releases no more.
Words in this milestoneIn order of appearance; open for the full explanation
More troubleshooting references
Stuck? Start with what actually happens
2 situationsThese paths synthesize official docs and public reports, not frequency data. Check the symptom and verify the result.
Tasks are too vague
Observed symptom: “Complete frontend” does not identify a next action or finish line.
- What should you check first?
Check each task for inputs, one output, and a verification action.
- Then resolve
Split work into independently demonstrable steps and list dependencies. Start only the first ready item.
- Verify the result
Each task names a visible result and pass criteria.
Edit the template with actual material, confirm, then copy into the current project conversation. Mark unknowns as undecided and replace examples as needed.
Confirmed · personalized content ready to copy
I am blocked at: Tasks are too vague.
Project and tool: [name, version, and project path]
Observed behavior: [steps, exact error, or screenshot; remove secrets and personal data]
Expected result: [actual goal]
First inspect: Check each task for inputs, one output, and a verification action.
Proposed approach: Split work into independently demonstrable steps and list dependencies. Start only the first ready item.
Explain the evidence and cause before making the smallest relevant fix. Preserve existing changes and do not add features. If I must act, give one precise action at a time and its expected visible result.
Success criteria: Each task names a visible result and pass criteria.
Report actual checks, unchecked items, and the next step.ReferencesGitHub · Planning and tracking work ↗
A new chat loses project progress
Observed symptom: AI repeats edits or skips unfinished tasks.
- What should you check first?
Read tasks/todo.md, Git status, and the last verification record.
- Then resolve
Record completed work, current work, blockers, next item, and relevant files. Restate this before continuing.
- Verify the result
A fresh chat resumes from recorded evidence.
Edit the template with actual material, confirm, then copy into the current project conversation. Mark unknowns as undecided and replace examples as needed.
Confirmed · personalized content ready to copy
I am blocked at: A new chat loses project progress.
Project and tool: [name, version, and project path]
Observed behavior: [steps, exact error, or screenshot; remove secrets and personal data]
Expected result: [actual goal]
First inspect: Read tasks/todo.md, Git status, and the last verification record.
Proposed approach: Record completed work, current work, blockers, next item, and relevant files. Restate this before continuing.
Explain the evidence and cause before making the smallest relevant fix. Preserve existing changes and do not add features. If I must act, give one precise action at a time and its expected visible result.
Success criteria: A fresh chat resumes from recorded evidence.
Report actual checks, unchecked items, and the next step.ReferencesGitHub · Planning and tracking work ↗Claude Code · Project instructions ↗
References for this milestone
Open these when the question arises; they are not extra required actions.
- How can AI stay on task?
When sending a new task, requesting an edit or reporting an error.
- How can success or failure be verified?
When a feature can be used or a repair needs retesting.
My project material and backups
Material stays in this browser, not in project files or AI chat. Chinese and English template drafts are separate; project selection and progress are shared. Do not enter passwords or API keys here.
Includes idea, progress, template drafts, personal checks and selection conditions, but not project code. Restore creates a new draft and preserves existing projects.
Download/import unavailable? Restore backup text
Draft and progress stay in this browser.
Complete the actions in order; expand the matching checks and remedies if you get stuck.