execute-fleet

Orchestrates parallel task execution across isolated worktrees with verified merges and evidence gates.

593|58|Updated Oct 21, 2025
One-click install
npx skills add https://github.com/anombyte93/prd-taskmaster --skill execute-fleet
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: execute-fleet
Source: https://github.com/anombyte93/prd-taskmaster/tree/main/skills/execute-fleet
Command: npx skills add https://github.com/anombyte93/prd-taskmaster --skill execute-fleet

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Coordinating multiple AI coding workers on a shared task graph often leads to conflicting edits, unverified completion claims, and broken integration branches. This Skill enforces a disciplined fleet execution model where workers build in isolated worktrees and tasks are only marked done after evidence-backed verification.

Core Features & Use Cases

  • Wave-based dispatch: Computes dependency-ordered execution waves from the task graph and spawns one worker per task chunk with explicit model routing from a cost-efficiency ladder.
  • Verified completion: Accepts only CDD cards and clean evidence files as proof of done, rejecting narration-only claims and enforcing a sole-writer rule for shared state files.
  • Sequential integration: Merges verified worker branches into a fleet-integration branch one at a time, running build gates after each merge and finishing with a ship-check before opening a final PR.
  • Use Case: A licensed Atlas Fleet run with 14 parsed tasks is dispatched across Claude, Codex, and Gemini workers in three waves, each branch verified and merged until the integration branch passes ship-check.

Quick Start

Run the execute-fleet phase for this project using the existing tasks.json and complexity report, dispatching workers in waves and merging verified branches into fleet-integration.

Frequently Asked Questions about execute-fleet

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run parallel AI coding workers on a task graph?

Use wave-based fleet execution: compute runnable task waves from the dependency graph, spawn one worker per task in an isolated git worktree, and collect terminal results through an inbox. Merge verified branches sequentially into an integration branch.

How does fleet execution verify that a task is actually done?

Completion is verified from the worker branch, not narration. The branch must contain a CDD card at .atlas-ai/cdd/task-<id>.json, evidence files with no non-zero exit status lines, and no edits to shared state files like tasks.json.

Can Claude, Codex, and Gemini workers run in the same fleet?

Yes. The wave computation returns a routing map assigning each task a backend and model from a capability ladder, so frontier models handle the hardest tasks while cheaper models take trivial ones. Non-Claude backends require experimental_backends enabled.

What happens when a fleet worker goes silent or fails?

A silent or dead worker is re-queued once with a fresh prompt. On a second failure the task is marked BLOCKED in the orchestrator scoreboard and execution continues with remaining independent tasks.

What are the prerequisites for starting a fleet run?

All hard gates must pass: premium tier capability detection, launcher inbox availability, existing tasks.json and task-complexity-report.json, a clean git status, and a defined integration branch policy. Fleet never falls back to solo execution.

When should I not use parallel fleet execution?

Avoid it without a licensed premium tier, a committed clean base, or a validated task graph. Merge conflicts are never force-resolved; they become separate fix tasks, so highly interdependent tasks may serialize anyway.