Generating Prd Json From Prd Md

PRD to JSON Conversion

Hybrid approach: Python script parses, AI validates and corrects.

Workflow

1. Dry-run parser (Bash tool) — catch parser issues before writing

python ralph/scripts/generate_prd_json.py --dry-run

Check output for: declared vs parsed story count mismatch, missing stories, empty acceptance/files. If issues found, fix PRD markdown or note for manual correction in step 3.

1. Run parser (Bash tool)

python ralph/scripts/generate_prd_json.py

Script handles: PRD.md parsing, (depends: ...) extraction, content hashing, state preservation.

1. Validate (Read tool)

• Read ralph/docs/prd.json (script output)
• Read docs/PRD.md (cross-reference)
• Check against Validation Checklist

1. Correct errors (Write tool, if needed)

• Fix issues found
• Recompute content_hash if title/description/acceptance changed
• Write corrected ralph/docs/prd.json

1. Report

• Story count and status
• Corrections made
• Suggest: make ralph_run

Validation Checklist

For each story, verify:

• [ ] id follows STORY-XXX format
• [ ] title is 3-7 words, matches PRD.md feature
• [ ] description is non-empty
• [ ] acceptance array is non-empty
• [ ] files array contains valid paths (if specified in PRD.md)
• [ ] content_hash is 64-char hex string
• [ ] depends_on references valid STORY-XXX IDs (no circular deps, no self-refs)

Cross-reference with PRD.md:

• [ ] All #### Feature N: headings have corresponding stories
• [ ] Story order matches PRD.md feature order
• [ ] (depends: STORY-XXX) syntax correctly parsed

Common Issues to Correct

| Issue | Correction | | ------- | ------------ | | Empty acceptance | Extract from description or PRD.md feature | | Invalid dependson reference | Remove non-existent story IDs | | Circular dependency | Remove one direction | | Missing contenthash | Recompute from title+description+acceptance | | Duplicate story IDs | Renumber sequentially |

prd.json Schema

See ralph/docs/templates/prd.json.template for structure and fields.

Usage

make ralph_prd_json

Next Steps

make ralph_init    # Validate environment
make ralph_run     # Start Ralph loop

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: qte77
• Source: qte77/claude-code-plugins
• License: Apache-2.0
• Homepage: https://qte77.github.io/claude-code-plugins/

Install and usage instructions live in the source repository linked above.