Spec Writer Skill

Goal

Transform a user goal or feature request into a structured PRD using EARS syntax. Before generating requirements, invoke the ambiguity layer to clarify unclear aspects. Human approval required before proceeding to implementation.

Boundary vs product/req-elicitation.skill: req-elicitation is the heavyweight IEEE 29148 / IREB elicitation pass (stakeholder interviews, use cases → full EARS_REQUIREMENTS.md). spec-writer is the fast feature-to-PRD path that reuses the same EARS forms (defined canonically in req-elicitation). For large or regulated requirement sets, run req-elicitation; use spec-writer for a single feature's PRD. Both must use the identical EARS patterns.

Steps

1. Receive and understand the goal

   • First, check for Step 0 strategic context in docs/strategic/ (e.g., BUSINESS_CASE.md, EMPATHY_MAP.md, PROBLEM_STATEMENT.md). If present, read them.
   • Ask the user: "Describe what you want to build in plain language"
   • Identify: Who is the user? What problem does it solve? What does success look like? (using the strategic docs if available)

2. Ambiguity Layer — ask clarifying questions Identify and ask about ALL ambiguous aspects before writing requirements:

   Questions to ask (examples):
   - "Who are the users of this feature? (roles, permissions, guest vs authenticated)"
   - "What happens if [edge case] occurs?"
   - "What are the performance expectations? (response time, concurrent users)"
   - "Is there an existing system this integrates with? What are its constraints?"
   - "What is explicitly out of scope for this version?"
   - "Are there any regulatory requirements? (GDPR, HIPAA, PCI-DSS)"
   - "What devices/browsers must be supported?"
   

   Wait for answers before proceeding.

3. Generate the PRD using templates/product/PRD_TEMPLATE.md

4. Write EARS requirements — use all 5 patterns as appropriate:

   Ubiquitous (always applies):

   THE SYSTEM SHALL encrypt all data in transit using TLS 1.3 or higher.
   

   Event-Driven (triggered by an event):

   WHEN a user submits a login form with incorrect credentials 3 consecutive times
   THE SYSTEM SHALL lock the account for 15 minutes and send an email notification.
   

   State-Driven (while in a state):

   WHILE the payment processing service is unavailable
   THE SYSTEM SHALL queue requests and retry with exponential backoff up to 3 times.
   

   Optional Feature (feature flag / conditional):

   WHERE two-factor authentication is enabled for the account
   THE SYSTEM SHALL require a TOTP code on each login attempt.
   

   Unwanted Behavior (defensive requirements — canonical IF…THEN form):

   IF a password is submitted for storage THEN THE SYSTEM SHALL hash it and SHALL NOT store the plaintext in any medium.
   IF an unhandled error occurs THEN THE SYSTEM SHALL return a generic error and SHALL NOT expose internal stack traces in HTTP responses.
   

   (Matches the IF [unwanted condition] THEN the [system] shall [protective response] form in req-elicitation.skill / req-review.skill.)

5. Write acceptance criteria Each requirement needs 1+ verifiable acceptance criteria:

   Given [precondition]
   When [action]
   Then [expected outcome with measurable result]
   

6. Non-functional requirements Always include:

   • Performance: expected response time, throughput
   • Security: data classification, access control requirements
   • Accessibility: WCAG level (if UI feature)
   • Scalability: expected load, growth projections

7. Out of scope section Explicitly list what this feature does NOT include. This prevents scope creep.

8. Present PRD to human for approval Say: "Here is the PRD. Please review and approve before I proceed to architecture/planning."

9. After approval: create SDLC checkpoint S01

   • Invoke sdlc-checkpoint.skill → creates SESSION_SNAPSHOT/S01_requirements.md
   • Updates PROJECT_STATE.md → current phase: S02_DESIGN

Constraints

• NEVER skip the ambiguity layer — unresolved ambiguity is the #1 cause of rework
• NEVER start planning or implementation without human approval of the PRD
• All requirements must be EARS-formatted and verifiable (not vague like "fast" or "user-friendly")
• Use the PRD template in templates/product/PRD_TEMPLATE.md

Output Format

Completed PRD.md file at project root (or docs/) using templates/product/PRD_TEMPLATE.md structure.

Security & Guardrails

1. Skill Security (Spec Writer)

• Mandatory Threat Modeling: The agent must intrinsically embed a lightweight threat model into the PRD generation process. If the feature involves PII, financial transactions, or external system integration, the agent must refuse to generate the final PRD until a Security NFR explicitly addresses Data-in-Transit and Data-at-Rest.
• Strict NFR Templating: The agent is barred from inventing subjective Non-Functional Requirements (e.g., "The system must be secure from hackers"). All security NFRs must map to verifiable industry standards (e.g., "The system shall comply with OWASP ASVS Level 2").

2. System Integration Security

• Explicit Out-of-Scope Privileges: The "Out of Scope" section must explicitly name prohibited downstream integrations or data flows. If building a caching service, it must explicitly state "This service SHALL NOT have long-term administrative database access."
• Data Classification Anchoring: Every PRD must start with a Data Classification directive (Public, Internal, Confidential, Restricted). The agent must use this anchor to determine the baseline cryptographic and authorization EARS requirements needed for the document.

3. LLM & Agent Guardrails

• Requirement Sanitization: The agent must scrub any user prompt inputs that attempt to explicitly define hardcoded passwords, developer backdoor accounts, or disabled SSL verifications before converting them into EARS requirements.
• Prompt Fatigue Vulnerability: A user might attempt to exhaust the agent's "Ambiguity Layer" by giving rapid, nonsensical answers. The agent must enforce a threshold (e.g., 3 failed clarification attempts) and cleanly abort the PRD generation rather than outputting a compromised or hallucinated specification.