| name | prd-parsing-logic |
| type | skill |
| description | Parses an Agentic PRD (knowledge/prd.md) to extract structured configuration for automated project generation. Extracts Epics, Stories (with JSON acceptance criteria), NFRs, and AI components. Use when initializing a project or feature implementation from a verified PRD. |
| license | MIT |
| allowed-tools | Read |
| metadata | {"version":"1.0.0","phase":"requirements","llm-preference":"claude","standard":"agentic-prd-parsing"} |
| agents | ["system-architecture-specialist"] |
| knowledge | ["none"] |
| templates | ["none"] |
| related_skills | ["writing-prd","automated-code-generation"] |
PRD Parsing Logic
Extracts machine-readable signals from the Factory standard PRD format.
Prerequisites
- A valid
knowledge/prd.md file following the Factory PRD template.
- Target PRD must be in
READY state.
Process
- Load PRD: Read
knowledge/prd.md.
- Identify JSON Blocks: Locate all ```json blocks within the "Functional Requirements" section. These contain machine-executable acceptance criteria.
- Extract Metadata:
- Extract
project_name from H1.
- Extract
description from Executive Summary.
- Extract AI Components: Parse Section 5 "AI Component Specifications" to identify model requirements and grounding strategies.
- Map to Config: Convert the extracted data into a
ProjectConfig schema.
Guidelines
- Regex Precision: Use robust regex to find the start and end of specific sections (Functional Requirements, NFRs, etc.).
- Validation: Ensure that every extracted JSON block is valid and follows the
story_id / acceptance structure.
- Error Handling: If a required section is missing, flag it as a warning but continue if possible.
Best Practices
- Schema Strictness: Validate extracted JSON blocks against
story_id and acceptance criteria schemas.
- Context Preservation: Ensure AI component specs include all necessary grounding and model parameters.
- Idempotency: Parsing the same PRD multiple times should yield identical
ProjectConfig objects.
When to Use
- When a
knowledge/prd.md exists and is marked as READY.
- Before running the
automated-code-generation skill.