/continue

Arguments: $ARGUMENTS

Goal: Resume implementation execution from last saved checkpoint.

Core Principles:

• Load checkpoint state to understand where to resume
• Handle failures gracefully with clear error messages
• Continue from exact point of interruption
• Update status as execution progresses

Phase 1: Load Checkpoint Goal: Read execution status and find resumption point

Actions:

• Parse $ARGUMENTS for spec name
• Set SPEC_NAME variable from first argument
• Check status file exists: !{bash test -f .claude/execution/$SPEC_NAME.json && echo "exists" || echo "missing"}
• If missing: Display "No execution to resume for $SPEC_NAME. Run /implementation:execute $SPEC_NAME to start fresh."
• Read status file: @.claude/execution/$SPEC_NAME.json
• Identify current state:
  • Last completed layer
  • Current incomplete layer
  • Failed tasks (if any)
  • Next task to execute
  • Total progress percentage

Phase 2: Determine Resumption Strategy Goal: Decide how to proceed based on checkpoint state

Actions:

• If status shows "failed" state:
  • Display error details from checkpoint
  • Ask user: "The previous execution failed. Have you fixed the issue? (yes/no)"
  • If yes: Mark failed task as ready for retry
  • If no: Display "Please fix the issue before continuing" and exit
• If status shows "paused" state:
  • Display "Resuming from paused state at Layer X, Task Y"
  • Identify next pending task in current layer
• If status shows "completed":
  • Display "Execution already complete for $SPEC_NAME"
  • Show summary and exit
• If no clear next task:
  • Display "Unable to determine resumption point. Run /implementation:execute $SPEC_NAME to restart."

Phase 3: Resume Execution Goal: Continue from checkpoint with same layer-by-layer approach

Actions:

• Load layered tasks: @specs/$SPEC_NAME/layered-tasks.md
• Load project context: @.claude/project.json
• Resume execution at identified resumption point:
  • If retrying failed task: Re-execute that specific task
  • If continuing layer: Execute remaining tasks in current layer
  • If starting new layer: Execute all tasks in next layer
• Follow same execution pattern as /implementation:execute:
  • Execute tasks in dependency order
  • Update status after each task completion
  • Write progress to .claude/execution/$SPEC_NAME.json
  • Handle errors and pause points
• Continue until layer completes or error occurs
• If all layers complete: Mark status as "completed"

Phase 4: Summary Goal: Report resumption results and current state

Actions:

• Display execution summary:
  • Resumed from: Layer X, Task Y
  • Tasks completed this session: N
  • Current status: Complete | In Progress | Failed
  • Overall progress: X% complete
• If execution complete:
  • Display "All layers executed successfully"
  • Next steps: "/quality:validate-code $SPEC_NAME to verify implementation"
• If execution incomplete:
  • Display "Execution paused or failed"
  • Next steps: "/implementation:continue $SPEC_NAME to resume after fixing issues"
• If execution failed:
  • Display error details
  • Next steps: "Fix the issue and run /implementation:continue $SPEC_NAME"