agentic-sdd-router

name	agentic-sdd-router
description	Route spec-driven development work to the correct phase skill. Use when the user asks for SDD, spec-driven development, Spec Kit style workflow, feature planning, implementation from specs, verification, or spec drift review.
allowed-tools	Read, Glob, Grep, Bash
risk	safe
source	local

Agentic SDD Router

Use this skill first for spec-driven development work. It chooses the current phase, loads only the needed phase skill, and prevents accidental jumps from vague intent to code.

Source Backing

This workflow is based on:

• Local guide: /Users/nawodyaishan/Downloads/production-spec-driven-agentic-coding-guide.md
• GitHub Spec Kit workflow: constitution, specify, clarify, plan, tasks, analyze/checklist, implement
• OpenAI Codex safety guidance: bounded sandboxing, explicit approvals for high-risk actions, and auditable tool activity

Phase Map

User intent	Use skill
Set up SDD in a repo	agentic-sdd-bootstrap
Write or improve product intent	agentic-sdd-spec
Resolve unclear requirements	agentic-sdd-spec
Create a technical plan	agentic-sdd-plan
Review plan before coding	agentic-sdd-architecture-review
Break plan into agent tasks	agentic-sdd-tasks
Implement an approved task	agentic-sdd-implement
Verify implementation or review diff	agentic-sdd-verification-review
Detect drift, release notes, retrospective	agentic-sdd-drift-retro

Required Order

For non-trivial production work, enforce this order:

bootstrap, if needed
-> spec
-> clarify, if needed
-> plan
-> architecture review and approval
-> tasks
-> implementation
-> verification and code review
-> drift check, release notes, retrospective

Routing Rules

• If there is no spec.md for the requested work, route to agentic-sdd-spec.
• If spec.md has open questions or ambiguous requirements, stay in agentic-sdd-spec.
• If there is no plan.md, route to agentic-sdd-plan.
• If plan.md exists but lacks approval, route to agentic-sdd-architecture-review.
• If there is no tasks.md, route to agentic-sdd-tasks.
• If implementation is requested and the plan is not approved, stop and ask for architecture approval.
• If implementation is requested for high-risk areas, require explicit approval before edits:
  • dependencies or lockfiles
  • database migrations
  • auth or authorization
  • payments
  • secrets or credentials
  • production config, infrastructure, deployment
• If the user asks for parallel agents or subagents, allow them only inside the implementation or review phases and only for bounded, disjoint work.

Repository Discovery

Before choosing a phase, inspect:

• .specify/memory/constitution.md
• docs/vision.md
• docs/adr/
• specs/
• AGENTS.md, CLAUDE.md, GEMINI.md
• package/build/test files for verification commands

Stop Conditions

Stop before implementation when:

• product behavior is unclear
• security or data sensitivity is unresolved
• the plan changes architecture without review
• migrations, auth, payments, secrets, dependencies, or production config are involved without approval
• verification commands cannot be identified

Output

Return the selected phase, the skill to use, missing artifacts, and the immediate next action.