Streams

What is a Stream?

A stream is a unit of parallel work within a plan. Each stream has:

• Title — what the stream does
• Description — detailed task specification for the agent
• File scope — glob patterns defining which files the stream can modify
• Dependencies — other streams that must complete before this one starts

Streams are the operator-facing execution unit. Each stream carries a task description derived from the discovery dossier. When a builder or reviewer encounters issues, Tack adjusts guidance for subsequent retry attempts within the same objective.

Parallel Execution

Streams without dependencies run in parallel. Tack schedules streams based on:

• Dependencies — stream B won't start until stream A completes
• Concurrency limit — configurable max concurrent agents

agents:
  max_concurrent: 4

File Scope

Each stream gets a file scope — a set of glob patterns that define its territory:

Stream 1: ["src/api/routes/**", "src/api/middleware/**"]
Stream 2: ["src/components/**", "src/hooks/**"]

File scope affects:

• Quality gates — gate failures are only attributed to the stream if errors are in its files
• Agent guidance — agents know their boundaries

Dependencies

Streams can depend on other streams:

Stream 1: "Add database migration" (no dependencies)
Stream 2: "Add API endpoint" (depends on Stream 1)
Stream 3: "Add frontend form" (depends on Stream 2)

Tack respects dependencies at both execution time (scheduling) and merge time (merge ordering).

Stream Lifecycle

See Also

• Quality Gates — how gate failures are attributed to streams
• Recovery & Retries — what happens when a stream fails
• Merge Queue — how merged streams are integrated
• Isolation & Sandboxes — how streams are isolated from each other