The planning-mode prompt for Step 2 of the “new project” workflow described in the README.
Paste it into your AI agent at the start of a new project, after Step 1 (set up the repo) and after you have a brief from Step 0. Adjust the bracketed parts.
I'm starting a new software project. Before we write any code or
file any tasks, I want you to use the AI development methodology
(located in this repo at docs/methodology/ — start with 00_README.md
for the mental model).
Here's what I know about the project so far:
- Product: [one or two sentences describing what you're building]
- Target user: [who it's for]
- Constraints: [budget, timeline, team size, tech preferences]
- What success looks like: [the outcome that would make this worth doing]
Your job, in order:
1. Read docs/methodology/00_README.md, 01_strategy.md, 02_pillars.md,
and 03_epics.md. Skim the others.
2. Ask me clarifying questions ONLY about things you genuinely cannot
guess. Don't ask permission questions; ask information questions.
3. Propose:
a) A draft strategy master plan at docs/strategy/00_master_plan.md
with vision, 3–5 phases (each with exit criteria), and the
document index. Follow the skeleton in 01_strategy.md.
b) An initial pillar set (5–8 pillars) as separate files at
docs/pillars/P1_*.md ... PN_*.md. Use the skeleton in
02_pillars.md. Sequential dependency chain.
c) The charter for the first epic at backlog/epics/E01-<slug>/README.md.
Use the template in 03_epics.md. One primary pillar. Binary
exit criteria. Folder naming uses the E<NN>-<slug> convention
per 03_epics.md "Standard epic-folder structure."
d) Project assurance baseline and accountable human per doc 07;
capability owners in the epic per doc 11. If implementers work
in parallel, declare both item WIP limits per doc 11.
e) The first 3–5 backlog items in that epic's BACKLOG.md. Use the
format in 04_backlog_items.md, including conditional assurance
fields, required reviewers, and evidence before implementation.
4. Use plan mode (your tool's planning feature). Show me each
artifact before moving to the next. Don't bundle.
5. Once everything is approved, fill in the starter ./CLAUDE.md
(already at the root from Step 1) with the project-specific
conventions, commands, and hard rules. The skeleton already
contains the working principles and a pointer to
docs/methodology/ as the authoritative source for HOW work is
done — you just need to fill in the <<placeholders>>.
The agent reads the methodology, asks the right questions, produces a complete planning skeleton. You review and refine. By the time you sit down to write code, you have strategy, pillars, an epic, and items — all in the right places.
For long-running autonomous milestone work (after the kickoff is done), see AUTONOMOUS_LOOP.md — a different prompt for a different mode of work.
For the ordinary phases after that — chartering the next epic, writing items, implementing, reviewing, verifying — see ROLE_BRIEFS.md.