Dev Workflow

Description

Standard R&D process orchestrator — guides the full development lifecycle from User Story to CD, with human confirmation required before proceeding to each step.

Collaboration with other skills: This skill is responsible for "what step are we on, what's next, and which skill to invoke"; architect handles "how to write the actual documents."

Every run must walk through the entire 9-step checklist — whether it is a new requirement, a new project, or a project already in progress — to ensure no critical step is missed (especially observability plans and test coverage).

Rules

1. Mandatory full checklist: Every run must go from Step 1 to Step 9; no step may be skipped, even if its status is "Completed"
2. Human confirmation at each step: After displaying the status of each step, wait for human confirmation ([y/N]) before proceeding to the next
3. Observability is not optional: Step 5 (observability plan) is a required step and must not be skipped with "not needed for now"
4. Documentation before code: Before Step 6 (TDD implementation), all documentation deliverables from Steps 1-5 must exist and be confirmed
5. First-run initialization: When the project has no CLAUDE.md or no workspace skill, automatically run project initialization
6. All User Stories must have E2E coverage: In Step 4's test plan, every User Story must have a corresponding L3 E2E test case

Examples

Bad — Skipping undiscovered missing steps

User: "Code is all done, MR is merged, help me set up CD"
AI: Directly discusses CD configuration...

Problem: Skipped the observability plan (Step 5), test plan check (Step 4), etc., jumping directly to the step the user mentioned.

Good — Enforced full checklist

User: "Code is all done, MR is merged, help me set up CD"
AI:
── Step 1 — User Story ──────────────
Status: Completed
Deliverable: docs/product/user-stories/postcard.md

Do you want to update this step? [y/N (default: skip)]

...(Steps 2-4 display status similarly)...

── Step 5 — Observability Plan ──────────────
Status: Missing
Deliverable: docs/architecture/postcard/observability.md — does not exist

Do you want to update this step? [y/N (default: skip)]

Even if the user says "code is done," the observability plan must still be checked for existence.

Bad — Triggering the full workflow for a simple bug fix

User: "Fix the login button color bug"
AI: Triggers dev-workflow, starts from Step 1 with User Story writing...

Problem: A single-line bug fix does not need the full R&D workflow; over-orchestration disrupts the user.

Good — Recognizing scope boundaries and avoiding over-triggering

User: "Fix the login button color bug"
AI: Directly fixes the bug following the project's Bug Fix TDD rules (write a failing test first, then fix).
Does not trigger dev-workflow.

Bad — Skipping Step 2.5 for UI features

User: "I want to build a new Postcard sharing page"
AI: After Step 2 tech design is done, jumps directly to Step 3 dev environment...

Problem: There is a UI change but the Paper design mockup is skipped; the visual approach is not aligned before development, risking repeated rework.

Good — Triggering Step 2.5 when there are UI changes

User: "I want to build a new Postcard sharing page"
AI: After Step 2 completes →
── Step 2.5 — UI Design Confirmation ──────────────
Status: Missing (detected that this requirement includes UI changes)
Trigger paper-ui-design skill to complete the design mockup in Paper...

First-Time Project Entry (Project Initialization)

Generate Project-Level Workspace Skill

Generate a project-specific development guide skill at .agents/skills/{project-name}-workspace/SKILL.md. This skill enables new Claude instances (or new engineers) to quickly understand all key project information.

Trigger condition: First time running dev-workflow in the project, or when .agents/skills/{project-name}-workspace/ does not exist.

Information sources: Collect from the following locations:

• docs/architecture/overview.md — System architecture
• docs/architecture/domain-model.md — Business terminology
• docs/architecture/tech-stack.md — Tech stack
• docs/deployment/local-dev.md — Local dev environment
• CLAUDE.md — Project rules
• Makefile or package.json — Common commands

Generated Workspace Skill structure:

---
name: {project-name}-workspace
description: {Project Name} project development guide. Automatically triggered when working in the {project-name} project.
  Contains project architecture, local environment startup, core business terms, common commands, and development conventions.
  New developers or Claude instances entering the project must invoke this skill first.
---

# {Project Name} Development Guide

## Project Overview
[Extracted from docs/architecture/overview.md]

## Quick Start
[Extracted make commands from docs/deployment/local-dev.md]

## Tech Stack
[Extracted from docs/architecture/tech-stack.md]

## Core Business Terms
[Extracted from docs/architecture/domain-model.md]

## Directory Structure
[Auto-scanned and generated]

## Development Standards
[Extracted from CLAUDE.md + dev-workflow rules]

## Common Commands Quick Reference
[Extracted from Makefile / package.json]

Location: .agents/skills/{project-name}-workspace/SKILL.md (project-level, committed to Git)

Update timing: After completing major documentation changes in Steps 1-4, update the workspace skill to keep it in sync.

Write the following base rules into the project's CLAUDE.md (skip if these rules already exist):

## Dev Workflow Rules

- All user stories MUST have E2E test coverage (L3 level). No User Story can be marked "implemented" without a corresponding passing E2E test.
- Follow the 9-step dev workflow: User Story → Tech Design → UI Design* → Dev Env* → Test Strategy → Observability → TDD Impl → MR → CI → CD (* conditional)
- Documents must be committed before coding starts (architect rule)

Full-Process Checklist (must be executed every time)

Whether it is a new requirement, a new project, or a project already in progress, every run must walk through Step 1 to Step 9 and confirm each step.

For each step, first scan the corresponding paths to determine the current status, then display the status and request confirmation:

Each step must receive human confirmation ("continue to next step?") before proceeding, even if the status is "Completed." This ensures every step receives conscious review rather than being automatically skipped.

Step 1 — User Story

Goal: Define user experience flows and acceptance criteria (no technical details).

Actions:

1. Ask the user if they have a Feishu document link. If so, invoke lark-mcp to read the document content as input.
2. Invoke story-craftsman for guided requirements discovery and AC definition.
3. If brainstorming is needed, invoke superpowers:brainstorming.
4. Write deliverables to docs/product/user-stories/{service}.md (following architect Step 1 standards).

Completion criteria: Each User Story has Background + User Story + AC (Given/When/Then), with no technical implementation details.

Write to CLAUDE.md after completion (if not already present):

- User stories location: docs/product/user-stories/
- Each User Story must have AC in Given/When/Then format

Confirmation: Display the User Story list and ask "User Stories complete, proceed to Step 2 (Tech Design)?"

Step 2 — Technical Design

Goal: Establish architecture boundaries, terminology SSOT, and key design decisions.

Actions:

1. Invoke superpowers:brainstorming to explore architecture options (2-3 options + recommendation).
2. Invoke superpowers:writing-plans to convert the plan into an implementation plan.
3. Write deliverables to docs/architecture/overview.md + docs/architecture/domain-model.md (following architect Step 2 standards).
4. Invoke doc-writing to refine documents using the HWPR/AWOR framework.

Completion criteria: Architecture diagram (Mermaid) + component responsibilities table + terminology mapping, single file ≤ 400 lines.

Confirmation: Ask "Tech design complete, proceed to Step 2.5 (UI Design Confirmation)?"

Step 2.5 — UI Design Confirmation (Conditional Trigger)

Trigger condition: This feature includes user interface changes (mobile pages, modals, cards, detail pages, etc.). If there are no UI changes, skip this step.

Goal: Before technical implementation, complete UI design in Paper based on User Stories and Design System Tokens, and output a component specification document to ensure the visual approach is aligned before development.

Actions:

1. Invoke paper-ui-design skill, completing the full Phase 1 + Phase 2 workflow:
   • Read relevant User Story documents and tokens from ~/.claude/skills/paper-ui-design/ds_token.md
   • Guide the user to open the Paper component library (optional skip)
   • Create a new Page in Paper and incrementally build the UI design mockup organized by visual groups
   • Take screenshots every 2-3 steps, evaluate against 7-item Review Checkpoint and fix issues
   • Call finish_working_on_nodes for finalization
2. After the user confirms the design in Paper, create or update docs/ui/ui-implementation-strategy.md in the project:
   • Add a separate section for this interface
   • List all components and controls, noting for each: UI component mapping, design system token correspondence
   • Follow the existing file format (if the file already exists, append the new interface section at the end; do not overwrite existing content)

Completion criteria:

• Paper design mockup is complete and passes all 7 Review Checkpoints
• docs/ui/ui-implementation-strategy.md has been created or updated with all component specifications for this UI

Confirmation: Ask "UI design confirmed, proceed to Step 3 (Dev Environment)?"

Step 3 — Dev/Test Environment (Conditional Trigger)

Trigger condition: The project has no existing dev/test environment, or new infrastructure dependencies (database, message queue, new service) or new third-party dependencies have been introduced.

If none of the above applies, skip this step.

Actions:

1. Invoke multi-worktree-dev to design the L1/L2/L3 layered local environment plan.
2. Write deliverables to docs/deployment/local-dev.md (following architect Step 3 standards).

Completion criteria: make dev-up / make dev-down are functional, and multiple worktree ports do not conflict.

Confirmation: Ask "Dev environment design complete, proceed to Step 4 (Test Plan)?"

Step 4 — Test Plan

Goal: Generate an L2-L4 layered testing strategy where all User Story acceptance criteria have corresponding test cases.

Actions:

1. Invoke testing-strategy to generate a complete layered plan based on project type (Backend+APP / Backend+WEB / Backend+APP+Embedded).
2. Confirm that every User Story AC has L3 (E2E) test case coverage (hard requirement).
3. Write deliverables to docs/testing/strategy.md and docs/testing/services/{service}/ (following architect Step 5 standards).

Completion criteria: L1 (unit) + L2 (integration) + L3 (E2E) + L4 (UAT) strategy is complete; User Story count = L3 test case count (1:1 correspondence).

Write to CLAUDE.md after completion (if not already present):

- L3 E2E tests are mandatory for every user story AC
- Test files location: docs/testing/

Confirmation: Ask "Test plan complete, proceed to Step 5 (Observability Plan)?"

Step 5 — Observability Plan

Goal: Before coding, define a measurement and observation system that is runnable, verifiable, and optimizable. Output a structured, executable, and traceable Observability Plan.

Execute the following 7 sub-steps in order; have the user confirm each sub-step's output before proceeding.

5.0 Scope & Classification

Read project information from docs/architecture/overview.md. If the file does not exist, infer first, then have the user confirm:

Output the scope confirmation table and wait for user confirmation before continuing.

5.1 Measurement Design

Define a measurement_id for this requirement. Format: {feature}_{action}_v{n}, e.g., postcard_share_activation_v1.

Guide the user through the following (provide suggestions based on understanding of the requirement first; user confirms or corrects):

Business objectives (based on the overall requirement, not individual User Story):

• What is the core product/business question?
• What are the key business KPIs?
• Define the critical path (for funnel definition)

If A/B experimentation is needed (based on the A/B experiment platform), additionally define:

Proceed after user confirms the Measurement Design.

5.2 Event Tracking Design

1. Confirm which new user behavior event tracking is needed for this User Story.
2. Present all candidate events in a table for user confirmation:

Each event must be evaluated as:

• reuse existing event — Fully reuse an existing event
• extend properties — Reuse the event but add new properties
• create new event — Create a new event only when necessary

Design each new/extended event structure according to the event tracking schema:

event_name:    xxx_yyy
classification: Impression / Behavior / Outcome
properties:
  - key: value_type  # Description
context:
  - user_id
  - session_id
  - feature_flag (if applicable)
ownership: Defaults to the user's team

After user confirms the event tracking plan, create a ticket in the event management platform for the review and release process.

5.3 Logging Strategy

Based on the requirement, define structured logs for critical business paths. Follow these principles:

• Only log critical business paths — use "event nodes" as the unit; not every function needs logging
• Use structured fields: service / trace_id / user_id / action / result
• Log levels: INFO (normal flow) / WARN (abnormal but recoverable) / ERROR (requires intervention)
• Correlation is required: trace_id + user_id
• Prohibited: Logging user privacy information (name, address, personal ID, etc.)

Output the log node list for user confirmation:

5.4 Tracing & Metrics

Based on the technical architecture, map the system's critical chains and design OTel instrumentation for key nodes/APIs:

Technical metrics (is the system running normally):

Business metrics (are KPIs being met):

Map the KPIs from 5.1 to computable metrics:

Output the plan for user confirmation.

5.5 Alerts & SLO

Define SLOs (internal targets) based on the system framework and set alert rules:

Severity levels:

• P0 Critical: Service unavailable / core feature down, immediate intervention
• P1 High: Severe impact but not fully unavailable, respond within 30 minutes
• P2 Medium: Issues but not urgent, handle during business hours
• P3 Low: Informational / observational

1. Invoke prometheus to review existing alert rules and confirm whether new PromQL alerts are needed.
2. Output the alert plan for user confirmation:

5.6 Dashboards

Invoke grafana to confirm whether new panels need to be added to dashboards:

5.7 Documentation Output

Write all deliverables from 5.0-5.6 to:

docs/architecture/{service}/observability.md

Document structure:

# Observability Plan — {feature} ({measurement_id})

## Scope & Classification
## Measurement Design
## Event Tracking
## Logging Strategy
## Tracing & Metrics
## Alerts & SLO
## Dashboards

Completion criteria:

• Measurement goal is clear; KPIs can be computed from the defined events
• Core funnel definition is complete
• Necessary events are designed (reuse preferred; do not proliferate new events)
• Critical paths have trace + metrics
• Core SLAs have alert coverage
• Support for experiment groups (if applicable)
• observability.md has been committed

Write to CLAUDE.md after completion (if not already present):

- Observability: event tracking + OTel + Prometheus alerts
- Observability docs: docs/architecture/{service}/observability.md

Confirmation: Ask "Observability plan complete, proceed to Step 6 (TDD Implementation)?"

Step 6 — TDD Implementation

Goal: Test-first implementation of all features, preferring Claude Code agent teams for parallel execution.

Actions:

1. Display the implementation plan (from Step 2's writing-plans deliverable) and ask the user:
   • Which tasks can run in parallel?
   • Which tasks must run sequentially?
2. Based on user confirmation:
   • Parallel tasks: invoke superpowers:dispatching-parallel-agents
   • Sequential / current session: invoke superpowers:subagent-driven-development
3. Each task follows superpowers:test-driven-development (write tests first, then implement).

Completion criteria: All User Story L3 E2E tests pass; L1/L2 tests pass; no broken tests.

Confirmation: Ask "Implementation complete, proceed to Step 7 (MR)?"

Step 7 — MR

Goal: Create a Merge Request and drive it to a mergeable state.

Actions:

1. Invoke gitlab-mr (auto-generates document links, pushes, creates MR, polls CI status, fixes failures).

Completion criteria: MR created successfully, CI all green, no merge conflicts.

Confirmation: Ask "MR created, proceed to Step 8 (CI Configuration)?"

Step 8 — CI

Goal: Configure or update the CI pipeline to ensure complete quality gates.

Actions:

1. Invoke gitlab-ci to check/create .gitlab-ci.yml.
2. Ensure the following CI stages exist: lint → test (including L1/L2) → build → E2E (L3).
3. Write deliverables to docs/deployment/ci.md (following architect Step 6 standards).

Completion criteria: CI pipeline all green; L3 E2E runs in CI.

Confirmation: Ask "CI configuration complete, proceed to Step 9 (CD)?"

Step 9 — CD

Goal: Configure continuous deployment to ensure automatic deployment to the target environment after MR merge.

Actions:

1. Invoke argocd to check current application deployment status.
2. Invoke cicd-developer to configure K8s/ArgoCD deployment strategy.
3. If a new cluster / full CICD stack is needed, invoke argocd-deploy.
4. Write deliverables to docs/deployment/cd.md (following architect Step 6 standards).

Completion criteria: Merge to main automatically triggers deployment; staging environment verification passes.

Write to CLAUDE.md after completion (if not already present):

- CD via ArgoCD, check docs/deployment/cd.md for deployment guide

Completion: Congratulations! The entire R&D workflow is complete. Ask the user if there are follow-up needs.

Flow Diagram

New feature ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  Step 1       Step 2       Step 2.5*    Step 3*      Step 4         Step 5          Step 6
  User Story → Tech     →  UI Design →  Dev Env   →  Test       →  Observability →  TDD Impl
  (story-      Design      (paper-ui-    (multi-       Strategy       (tracker-        (TDD +
  craftsman +  (brainstorm  design +     worktree-    (testing-       manager +        agents)
  lark-mcp)    + write-     docs/ui/)    dev)          strategy)      prometheus +
               plan +                                                 grafana)
               doc-writing)
                                              │
Re-entering ── status check ─────────────── resume from breakpoint ──────────────────────────────────────────────────────────

  Step 7       Step 8       Step 9
  MR        →  CI        →  CD
  (gitlab-mr)  (gitlab-ci)  (argocd +
                             cicd-developer)

* Step 2.5 only triggered when the feature includes UI changes
* Step 3 only triggered when there are new dependencies or no existing dev environment

Per-Step Confirmation Template

At the start of each step, display the current status; after completion, request confirmation. Always use the following format:

Entering each step:

── Step N/9 — [Step Name] ──────────────────
Status: Completed / Needs Update / Missing
Deliverable: [document path] — [one-line summary or "does not exist"]

Do you want to update this step? [y/N (default: skip)]

After completion or skip:

Step N confirmed → Proceeding to Step N+1 — [Step Name]

After all steps are complete:

══════════════════════════════════════
Full R&D Workflow Checklist Complete

Confirmed steps:
  Step 1   — User Story
  Step 2   — Technical Design
  Step 2.5 — UI Design Confirmation (with docs/ui/ui-implementation-strategy.md) [conditional]
  Step 3   — Dev Environment [conditional]
  Step 4   — Test Plan
  Step 5   — Observability Plan
  Step 6   — TDD Implementation
  Step 7   — MR
  Step 8   — CI
  Step 9   — CD
══════════════════════════════════════