// Generate an ML-centric PRD for ML-Ralph. Use when planning an ML project, experiment plan, or when asked to create an ML PRD. Triggers on: create a prd, write prd for, plan this ML feature, requirements for, spec out.
| name | prd |
| description | Generate an ML-centric PRD for ML-Ralph. Use when planning an ML project, experiment plan, or when asked to create an ML PRD. Triggers on: create a prd, write prd for, plan this ML feature, requirements for, spec out. |
Create ML-centric PRDs that are evidence-driven, stack-agnostic, and suitable for ML-Ralph.
tasks/prd-[feature-name].mdImportant: Do NOT start implementing. Just create the PRD.
Focus on:
Example format:
1) What is the primary objective?
A. Classification
B. Regression
C. Ranking
D. Other: [specify]
Generate the PRD with these sections:
Brief description of the ML task and why it matters.
Specific, measurable objectives (bullet list).
Explicit assumptions (data availability, metric definitions, constraints).
Each story must include:
Format:
### US-001: [Title]
**Description:** As a [role], I want [outcome] so that [benefit].
**Type:** discovery | experiment | evaluation | implementation | ops
**Hypothesis:** If ..., then ... because ...
**Evidence Required:** [What must appear in progress.txt or artifacts; include W&B run URL/ID for experiment/evaluation stories]
**Acceptance Criteria:**
- [ ] Specific, verifiable criterion
- [ ] Another criterion
- [ ] Ruff check passes
- [ ] Ruff format passes
- [ ] Mypy passes
- [ ] Pytest passes (if tests exist)
- [ ] Evidence logged in progress.txt
Important:
Numbered list of required behaviors or components.
Explicitly list what will not be done.
Known unknowns and how they’ll be resolved.
Define “done” in measurable terms.
Remaining questions that might alter the plan.
PRDs are living documents. ML-Ralph may refine prd.json each iteration based on evidence:
progress.txt.md)tasks/prd-[feature-name].md (kebab-case)