Implement Command

Execute tasks from the task list, respecting dependencies and updating progress.

Arguments

Input	Action
[ID]	Feature ID (optional if branch is associated)
(empty)	Next pending task
T001	Single task
T001-T005	Range of tasks
--all	All pending tasks

Arguments received: $ARGUMENTS

Process

Step 1: Resolve Feature

Parse arguments to extract:

• Feature ID (if provided as first numeric arg or XXX-name format)
• Task scope (T001, T001-T005, --all, or empty)

If no ID:

• Get current git branch
• Search .specs/*/spec.md for matching branch: in frontmatter
• If found, use that feature
• If not found:
  • If only one feature exists, use it
  • If multiple, list them and ask user to specify

Step 2: Load Context

Read from .specs/{ID}-{feature}/:

• spec.md - Requirements (extract Acceptance Criteria section)
• plan.md - Technical decisions and Critical Files
• tasks.md - Task list and progress

Check docs/research/ for any referenced research files.

If plan.md or tasks.md don't exist, inform user to run /spec-driven:plan and /spec-driven:tasks first.

Step 3: Update Status

If status is ready or draft:

• Update spec.md frontmatter to status: in-progress

Step 4: Load Critical Files

From plan.md, identify the ## Critical Files section.

For the tasks about to execute:

• Read the Reference Files relevant to current tasks (max 5 files)
• These provide patterns and conventions to follow

Step 5: Parse Scope

Determine which tasks to execute based on arguments.

Step 6: Validate Dependencies

For each task to execute:

• If marked [P], can proceed
• If marked [B:Txxx], check if Txxx is completed
• Skip blocked tasks, inform user

Step 7: Execute Tasks

Invoke the implement-agent with:

• Task scope
• Specification (Acceptance Criteria section)
• Technical plan
• Research summary (if exists)
• Current task list
• Reference file contents
• Feature ID and name

The agent will:

• Implement each task following the plan
• Validate against acceptance criteria
• Follow patterns from reference files
• Update tasks.md with checkboxes
• Suggest atomic commits

Step 8: Check Completion

After execution, check if ALL tasks are completed.

If all tasks done:

• Update spec.md frontmatter to status: to-review
• Inform user that implementation is complete

Step 9: Report

After execution:

• Show tasks completed
• Show files created/modified
• Show remaining tasks (if any)
• Suggest commit message
• If all done: suggest /spec-driven:validate for final review
• If tasks remain: suggest continuing with /spec-driven:implement

Error Handling

• Feature not found: List available features or suggest /spec-driven:init
• Plan/tasks not found: Inform user to run previous commands
• Dependency blocked: List which tasks need to complete first
• Implementation error: Report issue, keep task unchecked