repo-harness-prd

# repo-harness-prd Use this command to generate an upper-layer PRD under `plans/prds/`. The PRD is product intent and implementation guidance; it is not a Sprint backlog and does not start task execution. Activate `$geju` before drafting so the PRD starts from a high-altitude direction judgment. Prefer Claude CLI (`claude -p --model opus`) for PRD drafting; use Codex only as a fallback when Claude is unavailable, fails, or the user explicitly asks for Codex. ## Protocol 1. Confirm the working repo with `git rev-parse --show-toplevel`; read `docs/spec.md`, `.ai/harness/policy.json`, and the PRD template from `.prds.template_file` when present, otherwise `.claude/templates/prd.template.md`. 2. Accept a one-line or vague product idea and default to writing the PRD. Ask only when the answer would materially change platform, safety, legal risk, budget, data ownership, or scope tier. 3. Choose `compact` by default. Use `standard` only for multi-module products, explicit user request, commercialization, or frontend/backend deepening. 4. Activate `$geju` for a direction pass before PRD drafting. Produce a compact geju framing with Thesis, High-格局 Direction, Bold Takes, What Not To Do, First Proof Point, and Falsifier. This framing is input to the PRD, not a replacement for the PRD. 5. Apply the prior-art trigger rule before assembling the prompt: triggers are UI/taste decisions, market-convention patterns, library/framework selection, architecture precedent, or any `[UNVERIFIED]` ...