agentic-sdd-plan

name	agentic-sdd-plan
description	Produce a technical implementation plan from an approved SDD spec without editing product code.
allowed-tools	Read, Write, Edit, Glob, Grep, Bash, web_search_advanced_exa, web_fetch_exa
risk	safe
source	local

Agentic SDD Plan

Use this skill after the spec is clear enough to design the technical approach.

Goal

Create specs/<feature>/plan.md and related planning artifacts. Do not implement code in this phase.

Required Reading

• .specify/memory/constitution.md
• docs/vision.md
• specs/<feature>/spec.md
• specs/<feature>/clarify.md, if present
• relevant ADRs in docs/adr/
• relevant source files and tests
• existing API contracts, schemas, migrations, and deployment config when affected

Research Rules

Use Exa advanced search when:

• a framework, SDK, or platform behavior may have changed
• security guidance is current-version sensitive
• a public API contract is involved
• the plan depends on official docs

Prefer official docs, maintained repositories, standards, and primary sources. Record only facts that affect the plan.

Plan Sections

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

• summary
• inputs reviewed
• assumptions
• architecture approach
• affected modules
• API and contract changes
• data model changes
• dependency changes with alternatives and approval status
• security impact
• authorization boundaries
• observability impact
• testing strategy
• failure modes
• rollback and recovery
• risks and mitigations
• human architecture approval status

Create additional files when relevant:

• data-model.md
• contracts/openapi.yaml
• contracts/events.md
• test-plan.md
• ADR draft in docs/adr/

Architecture Rules

• Prefer existing project patterns and framework features.
• Avoid speculative abstractions.
• Justify every new dependency.
• Do not silently change auth, payments, data retention, or production operations.
• Include rollback for migrations and operationally risky changes.
• Include verification commands or observable results.

Stop Conditions

Stop and request human input when:

• dependency addition is proposed
• database migration is required
• auth, authorization, payments, secrets, or production config are touched
• rollback is not realistic
• tests or verification cannot be defined
• the plan conflicts with constitution or ADRs

Output

Return paths written, key decisions, risks, approval requirements, and readiness for agentic-sdd-architecture-review.