sdd-specify

SDD Specify

Use to turn an idea, feature, bug, or existing system into a behavior-first spec.

Artifact Shape

Prefer the repo shape: Spec Kit specs/<feature>/spec.md; Kiro .kiro/specs/<feature>/requirements.md or bugfix.md; OpenSpec openspec/specs/<capability>/spec.md plus openspec/changes/<change>/; fallback .sdd/<feature>/requirements.md.

Requirements Discipline

Keep specs at what/why level:

• source: user statement, code, runtime evidence, doc, regulation, or assumption
• users/goals/workflows and observable behavior
• acceptance, edge/failure cases, constraints, non-goals
• measurable success criteria

Avoid stack, classes, schemas, framework choices, and file paths unless bugfix/retrofit/design-first and explicitly labeled.

Ambiguity Policy

Proceed with labeled low-risk assumptions. Clarify, constrain, or convert to non-goal when ambiguity changes scope, security/privacy, UX, data retention, integration contracts, compliance, or delivery risk. [TO VERIFY] needs owner, risk, next action. LLM-drafted requirements are drafts until checked against user input, code truth, or project context.

Ask at most three concrete questions only when needed.

Required Content

Feature spec:

• problem, priority stories/journeys, non-goals
• requirement quality ledger: source, assumption status, ambiguity, reviewer/owner, evidence need
• stable functional IDs; measurable NFRs
• acceptance scenarios, edge/failure cases, assumptions/dependencies
• implementation-independent success criteria and readiness checklist

Bugfix spec:

• current, expected, unchanged behavior
• reproduction/evidence plus symptom source ledger
• suspected affected area, regression risk
• validation for original symptom and adjacent behavior

Retrofit spec:

• code/runtime behavior is truth
• inferred behavior is [TO VERIFY]
• separate current behavior from desired change
• record source per requirement: code, runtime proof, docs, or inference
• add initial "Spec Verification" task group
• do not invent schemas, endpoints, or deployment facts

Done

No placeholders; buildable requirements are testable or deferred; success criteria measurable; non-goals constrain scope; assumptions/sources/review gaps visible; stable IDs support traceability; next step is clarify, plan, tasks, or direct implementation.