IDEA → PROMPT → PLAN

Create a prompt

Describe your service. The prompt adds requests to clarify requirements, plan, and verify work. Review it, then copy it into your external AI tool.

kickoff does not provide an AI service.

JavaScript is off. Copy the base prompt below and append your service name, description, and considerations. Mark an undecided name as “Undecided” and blank considerations as “Needs clarification”. Paste the completed request into your external AI tool to discuss the design.Recommended: Start in Plan mode if your external AI tool supports it. This is optional.Read https://kickoff.jujin.dev/ai/startup/latest.md and https://kickoff.jujin.dev/ai/instructions.md, then begin project planning from my service requirements and constraints. Preserve confirmed decisions and existing authorization. Include version control systems and repository hosting in planning; preserve existing choices and resolve only what is still undecided. Use catalog articles only when I request them or they help explain a decision. Assess whether extra agent skills or workflow tools are useful, using uncertainty, dependencies, change risk, collaboration, handoff, existing practices, and setup cost. Compare starting without extra skills, Spec Kit, and suitable alternatives; explain fit and tradeoffs. After I accept or delegate the selection, install/configure needed components in my external agent within existing installation authorization, verify availability, and invoke the selected workflow through its supported mechanism. Reuse existing setup and records. If tools are unavailable, provide manual next steps and report what remains unverified. For unresolved user-owned choices, ask with realistic options, pros/cons, and a recommendation justified by project requirements and constraints. If I delegate with ‘use your recommendation’, decide and record within that context without asking again; continue within existing authorization and handle internal implementation yourself. Maintain a Markdown checklist of adopted project criteria; read relevant items before changes and verify actual results afterward, following the assistant rules. For larger tasks, maintain a task file, save each unit, verify it, and automatically commit locally after checks pass, subject to existing restrictions; follow the rules for remote actions or missing repository/write tools. On completion or interruption, report actual outcomes, applied design/architecture and key decisions with reasons and tradeoffs, user-confirmed versus delegated choices, checks, task-file location, save/commit status, and remaining work.

How to use →

Guidelines and versions

AI reference Markdown is in English. Write your own requests and project descriptions in your preferred language.

Read guidelines v1 Read Markdown (English) ↗AI behavior rules ↗

Guideline versions

v1 · revision 8 · 2026-10-08 — Fit-based agent workflows, starting without extra skills, and verified external-agent setup. History: v1 · revision 7 · 2026-10-01 — Markdown project checklists and relevant checks before and after changes. v1 · revision 6 · 2026-09-28 — Reasoned questions and scoped delegation, task records and local commits, and completion reports. v1 · revision 5 · 2026-09-27 — Version control systems and repository hosting in planning and prompts. v1 · revision 4 · 2026-09-26 — Proportionate architecture, SOLID/GRASP as judgment criteria, quality across software/design/monetization/operations, and maintenance handoff. v1 · revision 3 · 2026-09-24 — Two required documents, decisions from requirements, and optional catalog reading. v1 · revision 2 · 2026-09-23 — Project choices, AI-owned implementation and documentation, and operating constraints. v1 · revision 1 · 2026-09-22 — Initial release: context, catalog choices, software and design baselines, plan approval, and contributions.