flow-parallel

Decompose compound tasks into parallel work packages executed by independent claude -p processes.

Updated Aug 15, 2026
One-click install
npx skills add https://github.com/GongYuanCaiJi/dsh-claude-octopus --skill flow-parallel-gongyuancaiji
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: flow-parallel
Source: https://github.com/GongYuanCaiJi/dsh-claude-octopus/tree/main/.claude/skills/flow-parallel
Command: npx skills add https://github.com/GongYuanCaiJi/dsh-claude-octopus --skill flow-parallel-gongyuancaiji

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Large changes, migrations, and multi-issue fixes are too big for a single agent session, and naive parallelization causes overlapping edits and merge conflicts. This Skill decomposes a compound task into a validated Work Breakdown Structure and runs each work package as an isolated process with quality gates. ## Core Features & Use Cases - WBS Decomposition with Validation: Breaks a task into 3-10 non-overlapping work packages, validates the dependency graph for cycles, and assigns execution waves. - Isolated Parallel Execution: Each work package runs as an independent claude -p process in its own git worktree with explicit file-path instructions and credential isolation. - Wave-Based Orchestration: Dependent packages launch only after their dependency wave completes, with outputs injected into downstream instructions. - Use Case: Ask to build a full authentication system with OAuth, RBAC, and audit logging; the Skill splits it into three work packages, runs them in parallel, and aggregates results with quality spot-checks. ## Quick Start Ask the assistant to decompose and run a large task in parallel, for example: run /octo:parallel to build a full authentication system with OAuth, RBAC, and audit logging.

Frequently Asked Questions about flow-parallel

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

FAQPage Schema
How do I run a large coding task in parallel with Claude?▼

Invoke the parallel flow with a compound task description, and it decomposes the work into 3-10 work packages. Each package runs as an independent claude -p process in its own git worktree, with results aggregated after all waves complete.

How does parallel task decomposition avoid merge conflicts?▼

The WBS decomposition assigns explicit file paths and non-overlapping scopes to each work package, and an optional adversarial cross-check with a second model flags hidden dependencies or scope overlaps before launch. Each package also runs in an isolated git worktree.

Can work packages depend on each other in parallel execution?▼

Yes, packages can declare dependencies in the WBS. The dependency graph is validated for cycles, packages are assigned to waves, and outputs from completed dependency packages are injected into downstream instructions before their wave launches.

What happens when a parallel work package fails?▼

A failed package is flagged by its non-zero exit code, and its agent.log is surfaced for diagnosis. You can retry only the failed package individually without re-running the ones that succeeded.

What are the limitations of parallel agent execution?▼

The skill caps execution at 10 work packages because resource contention and merge complexity outweigh parallelism gains beyond that. It also requires the claude CLI to be available and cannot use Task tool subagents, which lack plugin access.