Add Agent Skill for AI-assisted PRD creation with structured interview process

[Copilot]

AI agents lack consolidated instructions for creating Ralph PRDs, leading to poorly structured task definitions and validation commands.

Changes

• New .github/copilot/agents/ralph-prd-creator/ agent skill following agentskills.io specification:

  • SKILL.md with YAML frontmatter (name, description, tools, license, compatibility)
  • Structured 4-phase interview process (460 lines):
    • Phase 1: Initial understanding (ask user to describe idea)
    • Phase 2: Discovery interview with 8 essential questions (ONE at a time)
    • Phase 3: Confirm understanding before proceeding
    • Phase 4: Clarifying questions with lettered options
  • Conversational guidelines for quality interviews:
    • Never ask multiple questions at once
    • Wait for answers between questions
    • Challenge vague responses
    • Push for specifics and concrete examples
  • 8-step task breakdown process with decision criteria
  • Task sizing rules (5-15min, single responsibility)
  • Acceptance criteria patterns (measurable vs subjective)
  • Priority-based dependency expression
  • Agent/model configuration (including cursor model requirement)
  • 4 templated patterns: Component, API, Utility, Bug Fix
  • DO/DON'T checklists and troubleshooting
  • references/prd-template.json - Complete PRD template

• Documentation updates:

  • Updated README to reference @ralph-prd-creator agent skill
  • Added "Agent Skills" section to main README listing all agent skills
  • Updated docs/README with agent skill reference
  • Updated /ralph:setup command to reference agent skill

Interview Methodology

Following the claude-build-workflow approach, the agent now conducts a structured discovery interview:

1. One question at a time - Never overwhelms the user
2. 8 essential questions - Problem, users, features, success, scope, integration, validation, tooling
3. Challenges vague answers - Gets specifics ("users can do things" → "what things? how?")
4. Confirms understanding - Summarizes before generating PRD
5. Quick decision making - Uses lettered options (1A, 2B, 3C)

This ensures high-quality PRDs that Ralph can execute successfully.

Agent Skills Format

Following the agentskills.io open standard specification provides:

• Standardized format with YAML metadata for agent discovery
• Tool compatibility declarations (bash, filesystem)
• Platform compatibility (claude, cursor, copilot, chatgpt)
• Portable structure with references/ subdirectory for templates
• Proper agent invocation via @ralph-prd-creator syntax

Example Pattern

The agent skill provides concrete templates like this component creation pattern:

{
  "id": "COMP-001",
  "title": "Create UserCard component",
  "acceptanceCriteria": [
    "File exists: src/components/UserCard.tsx",
    "Has props: { user: { name: string, email: string } }",
    "Displays user avatar (circular, 64px)",
    "typecheck passes",
    "test passes: pnpm test -- UserCard"
  ],
  "priority": 1,
  "passes": false,
  "notes": "Use Tailwind. Reference UserProfile.tsx for patterns."
}

vs. common mistakes (vague criteria like "UI looks professional" or "code is clean").

Usage

AI agents can now be invoked with @ralph-prd-creator to get expert PRD creation assistance with:

• Structured interview process
• Conversational guidance
• Templates and best practices
• All in one standardized skill package

Original prompt

---

This section details on the original issue you should resolve

<issue_title>feat: bundle better instructions for AI agents to create PRDs</issue_title>
<issue_description>Current AI agents have difficulty creating good PRDs because of how the system is designed. Including pre-made instructions on how to use this Ralph Coding Loop would help.</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes feat: bundle better instructions for AI agents to create PRDs #19

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[OctavianTocan]

@copilot, if you want to make a guide properly, talking to the AI, follow the agent skills format instead of just dumping a random Markdown file in the docs folder.

Here's what I mean: https://agentskills.io/specification

[OctavianTocan]

@copilot Here's more information on how the interview process should work: https://github.com/rohunj/claude-build-workflow

Pay attention to how they implement interviewing in that repo, and let's aim for at least that level.

[Copilot]

@copilot, if you want to make a guide properly, talking to the AI, follow the agent skills format instead of just dumping a random Markdown file in the docs folder.

Here's what I mean: https://agen...

Converted to Agent Skills format following the agentskills.io specification. Created .github/copilot/agents/ralph-prd-creator/ with:

• SKILL.md with proper YAML frontmatter (name, description, tools, license, compatibility)
• Structured instructions for AI agents
• references/prd-template.json for templates

The agent can now be invoked with @ralph-prd-creator syntax. (Commit: ef6c825)

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• agentskills.io
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node /home/REDACTED/work/_temp/ghcca-node/node/bin/node --enable-source-maps /home/REDACTED/work/_temp/copilot-developer-action-main/dist/index.js (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[Copilot]

@copilot Here's more information on how the interview process should work: https://github.com/rohunj/claude-build-workflow

Pay attention to how they implement interviewing in that repo, and let's ...

Enhanced the agent skill with a structured interview process following the claude-build-workflow methodology:

4-Phase Interview:

• Phase 1: Initial understanding (ask user to describe idea)
• Phase 2: Discovery interview with 8 essential questions (ONE at a time)
• Phase 3: Confirm understanding before proceeding
• Phase 4: Clarifying questions with lettered options

Conversational guidelines:

• Never ask multiple questions at once
• Wait for answers between questions
• Challenge vague responses
• Push for specifics ("users can do things" → "what things? how?")

The agent now conducts a conversational, guided interview that gets detailed requirements before generating PRDs. (Commit: 2970b48)