feat: Squad as Code — workflow orchestration DSL

[tt-a1i]

Summary

Define reusable multi-agent workflows in YAML and run them with a single command.

Usage

# .squad/workflows/full-feature.yaml
name: full-feature
steps:
  - agent: architect
    action: "Analyze requirements, output technical plan and file list"
    output_to: \$plan

  - parallel:
    - agent: worker-1
      action: "Implement backend based on \$plan"
    - agent: worker-2
      action: "Implement frontend based on \$plan"

  - agent: inspector
    action: "Review code from worker-1 and worker-2"
    on_fail: reassign worker-1

  - agent: writer
    action: "Generate docs and changelog from code changes"

squad run full-feature --goal "Add user login feature"

Features

• Sequential and parallel steps: Define execution order and parallelism
• Variable passing: Output from one step feeds into the next
• Conditional logic: on_fail, on_pass branching
• Reusable templates: Share workflows across projects
• Dry run: Preview the execution plan without running agents

Why

Repeated collaboration patterns should be codified, not manually orchestrated each time. This is the natural evolution of squad — from ad-hoc collaboration to reproducible workflows. Think GitHub Actions but for AI agents.

Complexity

Large. Requires a workflow engine, step executor, variable resolution, and error handling. This is a major feature that could define squad's long-term direction.