skills.rest
ledtorch
ledtorchCOMMUNITY
Explore all skills by this author →
Website𝕏@naivebaraGitHub

team

Orchestrates parallel sub-agent teams through a gated plan-run-review pipeline.

Updated Aug 10, 2026
Search Tags:#code review#workflow automation#claude code#codex#multi-agent#orchestration#task decomposition
One-click install
npx skills add https://github.com/ledtorch/coding-agent-workflow --skill team-ledtorch
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: team
Source: https://github.com/ledtorch/coding-agent-workflow/tree/main/skills/team
Command: npx skills add https://github.com/ledtorch/coding-agent-workflow --skill team-ledtorch

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Complex requests handled by a single agent often drift in scope, skip verification, or produce unreviewed output. This Skill decomposes a natural-language goal into a DAG of sub-agent tasks with two review gates, so planning and execution are checked before results reach the user. ## Core Features & Use Cases - DAG-based decomposition: Splits a goal into 3-5 typed nodes (Explore, Plan, general-purpose) dispatched in parallel layers, always ending with a critic node that must disagree. - Two review gates: Gate 1 classifies each plan finding as Auto or Ask before execution; Gate 2 validates each agent's output and re-dispatches only failed nodes, with a maximum of 2 rollbacks. - Hard resource caps: Enforces limits of 5 agents per layer, 10 agents per invocation, and 4 DAG layers to prevent over-fanning. - Use Case: Ask the agent to investigate and fix a cross-cutting bug; it reads the project reference docs, plans the investigation, dispatches parallel exploration agents, reviews their findings, and synthesizes a cited recommendation. ## Quick Start Run /team followed by a natural-language goal such as asking it to investigate why the login flow intermittently fails and propose a fix.

Frequently Asked Questions about team

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

FAQPage Schema
How do I run a multi-agent workflow in Claude Code?▼

Invoke /team followed by a natural-language goal. The skill decomposes it into 3-5 DAG nodes, dispatches dependency-free agents in parallel Task calls, reviews results at two gates, and synthesizes a final cited answer.

How does the team skill decide when to ask the user questions?▼

Gate 1 classifies each plan finding using Claude Code's auto-mode as the classifier: anything auto-mode would prompt on becomes an Ask, everything else is Auto. Ambiguous names or paths with multiple valid resolutions also become Ask.

Does the team skill work in Codex as well as Claude Code?▼

Yes. The same SKILL.md runs in both agents: invoke it as /team in Claude Code and as $team in Codex. The pipeline, gates, and caps behave identically in either environment.

What happens when a sub-agent fails or the plan is wrong?▼

Gate 2 re-dispatches only the failed node with a sharpened prompt, or rolls back to planning if the scope was wrong. After 2 rollbacks it stops and presents what it tried plus a next plan, waiting for user approval.

What are the limits of the team orchestration skill?▼

Hard caps allow at most 5 agents per layer, 10 agents per invocation, 4 DAG layers, and 2 rollbacks. Soft budgets suggest roughly 60k total tokens and 4k per agent report; exceeding them signals the task needs re-scoping.

Related Skills

Explore More Software Engineering Skills→
mkrlabsmkrlabs

openviking

Index and query semantic context from codebases via viking:// resources.

Official
Intermediate
bitsocialnetbitsocialnet
1

release-description

Generate a one-line release description from commits since the latest tag.

Official
Advanced
laicluselaicluse
4

drydry

Detects duplicated code, prose, and design patterns through scoped quick checks or full audits.

Official
Advanced
oskarroughoskarrough

arbe-orchestrate

Triages backlogs and design docs into dependency-wired dispatch plans for delegated workers.

Community
Advanced