AI Spec-Driven Development & Requirements Prompt Pack

Type "add CSV export to invoices" into a coding agent and you'll have working-looking code in minutes — plus a dozen decisions nobody made: which invoices, which month boundary, who may export, what happens with no data, what happens when two people click at once.

Most rework with AI coding agents comes from those hidden decisions: requirements invented to fill gaps, design choices slipped into requirements, acceptance criteria that can't pass or fail, long plans approved because they look thorough, "while I was there" refactors, specs that drift from the code, and "all tests pass" with no output to show for it.

AI Spec-Driven Development & Requirements Prompt Pack gives you 32 ready-to-paste prompts for Claude Code, Cursor, Codex, GitHub Copilot, Kiro, GitHub Spec Kit or any chat model that turn a vague request into a spec of what and why with stable requirement IDs, mark every gap [NEEDS CLARIFICATION] instead of guessing, write acceptance criteria you can test (Gherkin or EARS), produce a design and an agent-ready task list that point back at the spec, keep the agent inside scope with plan review and diff checks, and refuse "done" without evidence.

What you get

- PDF guide (59 pages, A4): why AI coding goes wrong without a spec, the project context block, "Tool facts in one page" (Spec Kit commands and gates, Kiro specs and steering, CLAUDE.md loading, Claude Code and Cursor plan mode, Cursor rules, Codex AGENTS.md discovery, Copilot instructions — checked against official docs, October 2026, with sources), a version snapshot, a "how much spec does this change need?" table, all 32 prompts with when-to-use notes, fictional examples of a spec.md, EARS requirements, Gherkin scenarios, a plan, a tasks.md, a single-task agent brief, an AGENTS.md section, a Cursor rule and a traceability matrix, reusable patterns, combinations for common situations, a fictional worked example, prompt anatomy and follow-up lines

- templates/: 32 plain-Markdown prompts + 3 checklists + README

The 32 prompts

- Foundations (01–04): project context block · request intake (problem vs solution) · "interview me" until the spec is clear · scope boundaries with MoSCoW and non-goals

- Requirements (05–09): spec / PRD draft (what & why, stable IDs) · user stories checked against INVEST · acceptance criteria in Gherkin · EARS requirements · non-functional requirements without borrowed numbers

- Clarify & review (10–13): ambiguity hunt · edge cases & states · spec review with a verdict · baseline spec from existing code (what must not change)

- Technical design (14–17): design doc tied to requirement IDs · design options compared honestly · exact data & interface deltas · rollout, risk & rollback plan

- Task breakdown (18–21): ordered task list with files, IDs and checks · single-task brief for a coding agent · parallel work plan for several agents · definition of done as commands

- Project rules & tool setup (22–25): project principles / constitution from evidence · spec-first section for AGENTS.md / CLAUDE.md / Cursor rules · map the workflow to Spec Kit, Kiro, Claude Code, Cursor, Codex or Copilot · steering / context files from the repo

- Plan-before-code & scope control (26–28): review an agent's plan before approval · scope-creep check on a diff · implementation checkpoint & handover

- Traceability, change & convergence (29–32): requirement → task → test matrix · test cases from acceptance criteria · change request (spec first, code last) · verify the implementation against the spec

Also included

- Spec quality checklist

- Plan & task readiness checklist

- Implementation, verification & change checklist

What's NOT included

- General agent setup (permissions, hooks, subagents, skills) and everyday agent prompts → AI Coding Agent Workflow Pack

- READMEs, ADR writing and reference docs → AI Documentation Prompt Pack

- Whole-system architecture → AI System Design & Architecture Prompt Pack

- OpenAPI / contract design → AI API Design & Contract Prompt Pack

- Writing the test code itself → AI Test Generation Prompt Pack

- Line-by-line PR review → AI Code Review Prompt Pack

- Fabricated benchmarks, productivity claims, testimonials or case studies — none appear in this pack

Accuracy

Tool behaviours in the guide were checked on 31 October 2026 against official documentation from GitHub Spec Kit, Kiro, Claude Code, Cursor, OpenAI Codex, GitHub Copilot and agents.md, plus the published sources for EARS, Gherkin, INVEST and MoSCoW. Versions quoted (specify-cli 1.2.5, Claude Code 2.1.316, Codex 0.166.0 and others) were the latest on npm, PyPI and GitHub on that date; prompts mark tool details VERIFY. All specs, task lists, rule files and the worked example are labelled fictional. Independent; not affiliated with or endorsed by GitHub, Microsoft, Amazon Web Services, Anthropic, Anysphere, OpenAI, Google or any vendor.

Price: $19

Author: Pluto · X @Mars_Peakul

Edition: 1.0 · October 2026