Developer Workflow

Developer Workflow

"This skill orchestrates the development workflow sequence. For detailed implementation guidance."

Core Responsibilities

• Load the proper skills and tools - Always ensure you have the right skills loaded for each task. Run tasks and subagents in parallel when possible to optimize time.
• Task Research - Research and plan implementation before coding. Read PRD, ask questions, research solutions, and create a clear implementation plan.
• Implementation - Write code to implement assigned tasks. Follow best practices, maintain code quality, and ensure functionality.
• Validation - Run validation steps and fix issues until passing. Use qa-validation-workflow skill for guidance.
  • CRITICAL: NEVER CREATE FAKE OR TRIVIAL TESTS. If a test cannot be created with meaningful assertions based on the specs/gdd requirements, DO NOT CREATE A TEST. Instead, report the issue to PM and document the gap in test coverage in the task comments and close the task as completed with observations.
• PRD Synchronization - Update PRD with implementation details, blockers, and observations. Keep PM informed of progress and issues.
• Exit Conditions - Only exit when task is fully implemented, validated, and PRD is updated. Never exit prematurely or leave tasks in an incomplete state.

Standard Message Pipeline

• Read message input in this order: CLI message argument → pending-messages-developer.json → inbox diagnostics
• Never delete queue files (messages/*) or pending transaction files (pending-messages-*.json)
• Watchdog owns delivery and cleanup lifecycle
• Signal lifecycle using status_update:
  • working when execution starts
  • awaiting_pm / awaiting_gd when blocked
  • ready / waiting / idle when available for next delivery

Agent Startup Protocol

On each Developer agent spawn:

1. Read Task Assignment
   • Check CLI message argument ($arguments.message for Claude CLI, or initial prompt for OpenCode CLI)
   • If CLI payload is empty, check ./.claude/session/pending-messages-developer.json
   • If both are missing, inspect ./.claude/session/messages/developer/*.json for diagnostics only
2. Read task state file read and understand current task details and status
3. Research task requirements Understand the requirements, read the specifications, use MCP tools (WebSearch, Fetch) to clarify implementation details, best practices, and potential blockers
4. Use your skills, tools, and subagents After understand the task requirements, review the available skills, subagents and tools, and activate the ones to use for the implementation.
5. Create a implementation plan document your approach in an implementation plan. Create steps and a checklist, update the task comments, and update PRD if needed
6. Update task state to "in_progress" in task state file (atomic write)
7. Process and implement the task following your plan and best practices, using the appropriate skills, subagents, and tools as needed
8. Run validation using qa-validation-workflow skill and fix any issues until passing
9. Run code base clean up subagent After any non-bug task completion, is mandatory to run the code-refactor subagent parallel task with the implemented context, to ensure code quality and maintainability
10. Update PRD and commit changes with implementation details, blockers, and observations (atomic write). Commit the changes following the default commit message pattern.
11. Notify the next agent wake up the next agent that needs to act after your actions via message system
12. Exit Send status_update (ready, waiting, or idle) to watchdog and wake up the next agent before exiting

IF BLOCKED

• Update state: state.status = "awaiting_pm", state.lastSeen = "{ISO_TIMESTAMP}"
• Document blocker in task prd.json
• Send message to PM with details
• Send status_update to watchdog with status: "awaiting_pm"
• Exit and wait

State Transitions

Converted and distributed by TomeVault — claim your Tome and manage your conversions.