gz-prd

Use to define or revise project-level intent as a PRD before ADR planning.

SKILL.md

GitHub

gz prd

Overview

Self-Escalation (opus-tier). The dialogue with the operator stays in the main session: a subagent cannot ask the operator a question or hear the answer, and what the operator adds is this skill's primary input. When the session model is below opus-tier, you may spawn an Agent with model="opus" for a bounded drafting or QC track that needs no operator input — pass the operator's words verbatim and the relevant context (ADR IDs, OBPI IDs, prior decisions), and treat what it returns as a draft you verify, not as the operator-facing result.

Scaffold a product requirements document, then fill it with the operator. gz prd <name> (prd in src/gzkit/commands/init_cmd.py):

  • requires an initialized project (.gzkit.json); otherwise it exits 1;
  • canonicalizes name to PRD-<SLUG>-<semver>: a leading PRD- is dropped, a trailing -X.Y.Z becomes the semver (default 1.0.0), and the rest is stripped to alphanumerics and upper-cased, so my-product becomes PRD-MYPRODUCT-1.0.0 (GHI #186). A name with no alphanumeric character exits 1;
  • renders the prd template (src/gzkit/templates/prd.md) with status: Draft, today's date, --title (default: the id) and the author prompts in AUTHOR_PROMPTS["prd"] (src/gzkit/templates/author_prompts.py) into <paths.prd>/<id>.md;
  • appends one prd_created ledger event, which makes the id a node in the gz state graph.

The author prompts are questions, not content: Problem Statement, North Star, Invariants and the Q&A Transcript stay unwritten until the operator answers them. It does not check for an existing file: a name that canonicalizes to an existing id overwrites that file and appends a second event. --dry-run prints the path and event it would write, whether or not the file exists, and writes nothing. gz interview prd is the interactive alternative: it asks the questions in PRD_QUESTIONS (src/gzkit/interview.py) and writes the answers into the same template.

Workflow

  1. Preview: uv run gz prd <name> --dry-run. If the printed path already exists, stop — that PRD is revised in place, not re-scaffolded.
  2. Create it: uv run gz prd <name> [--title "<title>"].
  3. Interview the operator for each author prompt and replace it with their answer; record the exchange under Q&A Transcript. required_headers in src/gzkit/schemas/prd.json is the authority for the sections and the permitted status values.
  4. Validate: uv run gz validate --documents checks the frontmatter and headers against that schema. Capture the output to a file and read $? straight after it.
  5. Report the id, path and validation result. ADRs are then planned against the PRD with gz-plan (uv run gz plan create).

Example

uv run gz prd my-product --dry-run
uv run gz prd my-product --title "My Product Requirements"
uv run gz validate --documents