parallel-workflows

Design and launch a dynamic workflow for a complex task. Claude writes a JavaScript orchestration script, fans it out across parallel agents, and synthesizes results — compressing sequential hours into minutes.

Fan out N specialized sub-agents across a codebase or file list simultaneously. Each sub-agent works in an isolated context window, then findings are adversarially verified and synthesized into a single report.

Review a Claude Code workflow script for correctness, pattern misuse, and quality issues. Reads the script at the given path, delegates critique to the workflow-architect agent across six dimensions, and returns a structured findings list with severity and suggested fixes.

Specialist in configuring, scoping, and prompting Claude Code sub-agents for use inside dynamic workflows. Use when the user needs help defining sub-agent system prompts, choosing tool access, selecting models, or structuring the prompt a workflow agent() call should send.

Expert at designing Claude Code dynamic workflow scripts. Use when the user needs to build, debug, or optimize a workflow orchestration script — especially when choosing between pipeline vs parallel, designing phase structure, or implementing adversarial verification patterns.

Use this skill when the user wants to understand, trigger, write, manage, or debug a Claude Code dynamic workflow. Covers when workflows beat subagents, how to trigger them, what the script primitives do, how to save and resume runs, and cost management.

Use this skill when writing or reviewing workflow scripts that need to handle agent failures, design defensive schemas, or correctly place .filter(Boolean) after parallel and pipeline calls. Reference whenever a workflow will run at scale where individual agent failures are expected.

Use this skill when the user wants to run parallel Claude sessions from the shell using `claude -p` without the full workflow runtime — for scripting, CI/CD pipelines, or cases where a lightweight fan-out is enough.

Use this skill when writing or reviewing a workflow script and needing to pick the right orchestration pattern — adversarial verify, loop-until-dry, judge panel, multi-modal sweep, or completeness critic. Provides copy-paste patterns with explanations of when each applies.