Skip to content
Arness

Batch Pipeline

When you have a backlog of features ready to build, the batch pipeline plans and implements them in parallel — each feature in its own isolated Git worktree, each with its own PR. Inspired by the recently announced Claude Code batch capabilities, Arness extends the concept with batch-merge for automated PR conflict resolution and batch-simplify for cross-feature deduplication after merge.

When to Use Batch

Section titled “When to Use Batch”
  • You have multiple features specced and planned (or ready to be planned)
  • Features are independent enough to implement in parallel
  • You want to maximize throughput without manual coordination

The Four Stages

Section titled “The Four Stages”

1. Batch Planning

Section titled “1. Batch Planning”
/arn-code-batch-planning

Batch planning takes multiple features and specs them in parallel:

  • Selects unblocked features from your backlog (Spark feature files, GitHub issues, Jira issues, or plain descriptions)
  • Spawns a batch-analyzer agent per feature to generate draft specifications
  • Reviews and refines specs through the standard feature-spec flow
  • Generates implementation plans for approved specs

Features with dependencies are sequenced automatically — independent features proceed in parallel.

2. Batch Implementation

Section titled “2. Batch Implementation”
/arn-code-batch-implement

Each planned feature gets its own background agent running in a separate Git worktree:

  • Worktree isolation — each agent works on an independent copy of the repository, preventing conflicts during development
  • Full pipeline per feature — each agent runs the standard execute-plan workflow (task execution, review, simplification)
  • Background execution — agents run concurrently while you continue working
  • PR per feature — each agent creates its own pull request when done

3. Batch Merge

Section titled “3. Batch Merge”
/arn-code-batch-merge

When implementation PRs are ready:

  • Discovers all open batch PRs
  • Analyzes for cross-feature conflicts using a batch-PR-analyzer agent
  • Determines the optimal merge order based on dependencies and conflict risk
  • Executes merges sequentially, resolving conflicts as needed

4. Batch Simplify

Section titled “4. Batch Simplify”
/arn-code-batch-simplify

After merging multiple features:

  • Scans for duplication introduced across features (two features may have created similar utilities)
  • Identifies consolidation opportunities
  • Proposes cross-feature refactoring to reduce redundancy

Example Workflow

Section titled “Example Workflow”

Starting with a Spark feature backlog containing 5 features:

/arn-code-batch-planning          # Specs and plans all 5 features
                                   # 3 are independent, 2 have dependencies


/arn-code-batch-implement          # 3 independent features build in parallel
                                   # 2 dependent features queue behind their blockers


/arn-code-batch-merge              # Merges all 5 PRs in dependency order


/arn-code-batch-simplify           # Finds and consolidates cross-feature duplication

Tips

Section titled “Tips”
  • Feature independence matters — the more independent your features are, the more parallelism you get. Spark’s feature extraction produces dependency-tracked features by design.
  • Review PRs before batch merge — each feature PR can be reviewed individually before merging the batch.
  • Works with any feature source — Spark feature files, GitHub issues, Jira issues, or plain descriptions all work as batch inputs.

Next Steps

Section titled “Next Steps”