agentic-sdd-spec

name	agentic-sdd-spec
description	Create or refine SDD feature, bug-fix, refactor, or migration specifications with clarification handling before technical planning.
allowed-tools	Read, Write, Edit, Glob, Grep, Bash, web_search_advanced_exa, web_fetch_exa
risk	safe
source	local

Agentic SDD Spec

Use this skill when creating or improving the product-facing specification for a change.

Goal

Capture what and why before deciding how. Do not design implementation in this phase.

Inputs

Read:

• docs/vision.md, if present
• .specify/memory/constitution.md, if present
• existing specs/*/spec.md
• relevant README/product docs
• user request

Use Exa only when the spec depends on current external facts, official product behavior, regulations, API behavior, or tool versions.

Spec Types

Choose one:

• feature spec
• bug fix spec
• refactor spec
• migration spec
• release/documentation spec

Feature Spec Sections

Create or update specs/<nnn-short-name>/spec.md with:

• problem statement
• goals
• non-goals
• users or actors
• user journeys
• functional requirements
• acceptance criteria
• success criteria
• edge cases
• data sensitivity and compliance notes
• API or integration expectations, without implementation design
• assumptions
• open questions
• human approval status

Clarification Rules

• Ask at most 3 clarification questions at a time.
• Ask only when the answer changes scope, security/privacy, user experience, data model, compatibility, or acceptance criteria.
• Use reasonable defaults for low-risk missing details and document them as assumptions.
• Do not proceed to planning while unresolved open questions affect implementation.

Quality Checklist

Create or update specs/<feature>/checklists/requirements.md with:

• no implementation details in spec
• goals and non-goals clear
• requirements testable and unambiguous
• acceptance criteria complete
• success criteria measurable and technology-agnostic
• edge cases identified
• data sensitivity noted
• open questions resolved or explicitly deferred
• spec ready for planning

Stop Conditions

Stop before planning if:

• user-visible behavior is ambiguous
• actors or permissions are unclear
• data sensitivity is unresolved
• success criteria are not verifiable
• the user has not approved or accepted critical clarifications

Output

Return the spec path, checklist path, unresolved questions, assumptions, and readiness for agentic-sdd-plan.