Skip to content
Back to skills

Generating Prd Json From Prd Md

ASecurity

Generates prd.json task tracking file from PRD.md requirements document. Use when initializing Ralph loop or when the user asks to convert PRD to JSON format for autonomous execution.

  • 2 stars
  • 0 votes
  • 1 copy
  • 1 view
  • Added September 5, 2026
ai-agentspythonbash

Works with

  • claude code

Security analysis

A100/100

Scanned September 5, 2026

npx -y skills add qte77/claude-code-plugins --skill generating-prd-json-from-prd-md --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Generating Prd Json From Prd Md?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Generating Prd Json From Prd Md
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/qte77-generating-prd-json-from-prd-md/badge)](https://www.skillsdirectory.com/skills/qte77-generating-prd-json-from-prd-md)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: generating-prd-json-from-prd-md
description: Generates prd.json task tracking file from PRD.md requirements document. Use when initializing Ralph loop or when the user asks to convert PRD to JSON format for autonomous execution.
compatibility: Designed for Claude Code
metadata:
  allowed-tools: Read, Write, Bash
  model: haiku
  stability: development
---

# 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

```bash
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)

```bash
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

2. **Correct errors** (Write tool, if needed)
   - Fix issues found
   - Recompute `content_hash` if title/description/acceptance changed
   - Write corrected `ralph/docs/prd.json`

3. **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 depends_on reference | Remove non-existent story IDs |
| Circular dependency | Remove one direction |
| Missing content_hash | 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

```bash
make ralph_prd_json
```

## Next Steps

```bash
make ralph_init    # Validate environment
make ralph_run     # Start Ralph loop
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.