Skip to content

implement.md

Latest commit

 

History

History
55 lines (36 loc) · 3.12 KB

implement.md

File metadata and controls

55 lines (36 loc) · 3.12 KB

/awos:implement

Runs tasks — delegates coding to sub-agents, tracks progress.

What it does

This command acts as an engineering manager — it reads the task list, delegates each task to the appropriate specialist subagent, and tracks progress. It does not write code itself. It updates:

  • context/spec/[index]-[name]/tasks.md — marks tasks [x] as they complete.

The actual code changes are made by the specialist subagents it delegates to.

Prerequisites

  • The target spec directory must contain functional-spec.md, technical-considerations.md, and tasks.md.
  • At least one incomplete task ([ ]) must exist in tasks.md.

How it works

  1. Finds the next task: Scans spec directories in order, finds the first incomplete task. Or uses your prompt to target a specific spec/task.
  2. Loads full context: Reads all three spec files (functional, technical, tasks) so the delegated agent has complete information.
  3. Extracts agent assignment: Reads the **[Agent: agent-name]** tag from the task description to determine which specialist to delegate to.
  4. Delegates to subagent: Sends a detailed prompt with full context and clear success criteria to the specialist agent (or general-purpose if no assignment found).
  5. Updates progress: Marks the completed task [x] in tasks.md. If all tasks under a slice are done, marks the slice header too.
  6. Reports status: Shows completion percentage (e.g., "5/12 tasks done (42%)").

Key behaviors

  • Strictly an orchestrator. The implement command is prohibited from writing, editing, or modifying any production code, configuration files, or database schemas. It only delegates and tracks.
  • Full context per delegation. Each subagent receives the complete functional spec, technical spec, and task list — not just the task description. This ensures agents have all the context they need.
  • Automatic task progression. When run without arguments, it automatically finds and starts the next incomplete task.
  • Slice/task tracking. When all tasks under a slice are marked complete, the slice header is automatically marked complete too.

Common misconceptions

  • "This command writes the code." No. It delegates to specialist subagents. The implement command is purely an orchestrator.
  • "I need to babysit every task." You don't. Run /awos:implement all the slices in the spec and let it work through the entire list autonomously. Each slice has clear scope, verification criteria, and agent assignments — the agents run the code and check results themselves.
  • "I should micromanage the agents." Trust the flow. Vertical slicing ensures each task is small and self-contained, and Claude Code can actually run the changes and verify that each slice meets its acceptance criteria.

Example usage

# Good — auto-pick the next task:
> /awos:implement

# Good — implement all remaining tasks:
> /awos:implement Implement all tasks

# Good — target a specific phase:
> /awos:implement Implement Phase 2

What happens next

When all tasks reach 100%, run /awos:verify to check acceptance criteria and mark the spec as completed.