Agent-Skills.md

Agent Skills: Feature Specification Skill

Guide iterative specification and CDC (Cahier Des Charges) creation through deep questioning, context analysis, and proactive proposals. Use this skill BEFORE feature-research to clarify requirements, identify prerequisites, define scope, and document complete specifications. Triggers when starting a new feature, task, bug fix, or refactoring and requirements need clarification.

UncategorizedID: LaizyIO/WorkflowSkills/feature-specification

Repository

LaizyIO/WorkflowSkills
LaizyIO

Install this agent skill to your local

pnpm dlx add-skill https://github.com/LaizyIO/WorkflowSkills/tree/HEAD/feature-specification

Skill Files

Browse the full folder contents for feature-specification.

Download Skill

Loading file tree…

feature-specification/SKILL.md

Skill Metadata

Name
feature-specification
Description
Guide iterative specification and CDC (Cahier Des Charges) creation through deep questioning, context analysis, and proactive proposals. Use this skill BEFORE feature-research to clarify requirements, identify prerequisites, define scope, and document complete specifications. Triggers when starting a new feature, task, bug fix, or refactoring and requirements need clarification.

Feature Specification Skill

Purpose

Transform vague or incomplete requests into comprehensive, validated specifications through iterative dialogue. This skill produces a CDC (Cahier Des Charges) document in French that serves as the foundation for all subsequent phases.

Key difference from feature-research:

  • feature-specification = WHAT to do (define precisely)
  • feature-research = HOW to do it (technical research)

IMPORTANT: User Interaction

ALWAYS use the AskUserQuestion tool to ask questions to the user.

This tool allows structured questioning with multiple choice options:

  • Ask 1-4 questions at a time
  • Provide 2-4 options per question with descriptions
  • User can select options or provide custom "Other" response
  • Use multiSelect: true when multiple answers are valid

Example usage:

AskUserQuestion:
  questions:
    - question: "Who are the primary users of this feature?"
      header: "Users"
      options:
        - label: "Internal employees"
          description: "Staff members using internal tools"
        - label: "External customers"
          description: "End users of the product"
        - label: "Administrators"
          description: "Users with elevated privileges"
      multiSelect: true

Why this matters:

  • Structured questions are clearer for users
  • Options help users think through possibilities
  • Responses are unambiguous
  • Conversation stays focused

Specification Workflow

Phase 1: Initial Context Analysis

  1. Read the user request - Understand the initial ask
  2. Analyze project context - Use Glob/Grep to discover:
    • Project structure and technologies
    • Existing similar features
    • Conventions and patterns in use
  3. Identify domain - Understand the business context
  4. Ask initial framing questions - Start the dialogue
Example initial analysis:
- Found: package.json → Node.js/React project
- Found: src/modules/auth/ → Authentication module exists
- Found: PostgreSQL migrations → Database with existing schema

Phase 2: Deep Iterative Questioning

Ask questions across these categories. See references/questioning-guide.md for complete guide.

2.1 Motivation & Objectives

  • What problem does this solve?
  • What are the business goals?
  • How will success be measured?

2.2 Users & Personas

  • Who will use this?
  • What is their technical level?
  • What are their current workflows?

2.3 Detailed Functionality

  • What exactly should happen?
  • What actions must be possible?
  • What are expected inputs/outputs?

2.4 Constraints & Limits

  • Technical constraints (performance, security)?
  • Business constraints (regulations, processes)?
  • Time or budget constraints?

2.5 Integration & Dependencies

  • What existing systems to integrate with?
  • What prerequisites must be in place?
  • Impact on existing functionality?

2.6 Edge Cases & Exceptions

  • What if X fails?
  • How to handle edge cases?
  • What error scenarios to anticipate?

Questioning principles:

  • Ask 2-4 questions at a time, not overwhelming lists
  • Build on previous answers
  • Validate understanding before moving on
  • Be specific based on project context discovered

Phase 3: Contextual Codebase Research

Explore the codebase to inform specifications:

  1. Find similar implementations

    Glob: **/*Service.ts, **/*Controller.cs
Grep: "similar-feature-name"

  2. Understand existing patterns

    • How are similar features structured?
    • What conventions are followed?
    • What dependencies are used?

  3. Identify integration points

    • Which components will be affected?
    • What interfaces exist?
    • What data models are relevant?

Phase 4: Proactive Proposals

Be a force of proposal, not just passive. See references/proposal-techniques.md.

Propose simplifications:

"Rather than a complete notification system, I suggest starting with
emails only. This validates the concept before adding push/SMS."

Suggest alternatives:

"I noticed you don't have a queue system. For emails, I suggest using
Hangfire which is already configured in the project."

Identify quick wins:

"We could reuse the existing EmailTemplate component from the
password-reset feature."

Anticipate problems:

"If we store files locally, this won't work with multiple server
instances. Consider using Azure Blob Storage instead."

Phase 5: CDC Documentation

Compile all gathered information into a CDC document.

  1. Create CDC.md using template from references/cdc-template.md
  2. Review completeness - Ensure all sections are filled appropriately
  3. Present for user approval

Context-Aware Questions

Adapt questions based on discovered project context:

If PostgreSQL detected:

"This feature will need new database tables. Have you thought about the data model?"

If React + TypeScript detected:

"Should we include E2E tests with Playwright for this feature?"

If similar feature found:

"I found a similar feature in src/modules/orders. Should we follow the same pattern?"

If authentication exists:

"Should this feature be restricted to authenticated users? Specific roles?"

Output

The skill produces CDC.md containing:

  • Clear problem statement and context
  • Validated objectives with success criteria
  • Explicit scope (in/out)
  • Functional requirements with acceptance criteria
  • Non-functional requirements (performance, security, accessibility)
  • Constraints and prerequisites
  • User personas and usage scenarios
  • Identified risks with mitigations
  • Documented decisions
  • User approval

See references/cdc-template.md for the complete template.

Tips for Effective Specification

  1. Ask before assuming - Clarify ambiguities early
  2. Be context-aware - Use project analysis to ask relevant questions
  3. Propose, don't just ask - Offer options based on your analysis
  4. Document as you go - Capture decisions in real-time
  5. Validate incrementally - Confirm understanding before moving on
  6. Think integration - Consider how this fits the existing system
  7. Anticipate problems - Surface issues early
  8. Keep scope focused - Help user avoid scope creep

Integration with Workflow

Phase 0: Specification (this skill)
   ↓ CDC.md
Phase 1: Research (feature-research)
   ↓ findings.md
Phase 2: Planning (implementation-planner)
   ...

The CDC becomes input for feature-research, providing clear requirements for technical research.

Bundled Resources

  • references/cdc-template.md - Complete CDC template in French
  • references/questioning-guide.md - Detailed questioning strategies by category
  • references/proposal-techniques.md - Techniques for being proactive