agentleFS
MGeorge98MGeorge98/claude-skills/orchestrator/SKILL.md

orchestrator

Skill0 starsChanged 39 days agoMITView on GitHub

CEO/orchestrator mode — the main session only plans, decides, and delegates; all real work goes to subagents (Sonnet for simple, Opus for hard) to keep the main context cheap. Use when the user says "orchestrator", "esti orchestratorul", "act like a CEO", "tu esti creierul", "deleaga", "trimite agenti", "orchestrator mode", or asks to minimize token cost by delegating work to agents.

88 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use worker, reviewer and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • stotihvSkill

    worker

    Execute beads autonomously within a track. Handles bead-to-bead context persistence via Agent Mail, uses preferred tools from AGENTS.md, and reports progress to orchestrator.

    Also in 3 projects that use this

  • MicrosoftSkill

    reviewer

    ASP.NET Core ハンドブックの記事をレビューする。Use when: 記事をレビューする、ドキュメントを校正する、コードレビューする、PRをレビューする。技術ドキュメントのレビュー、品質チェック、校正。

    Also in 4 projects that use this

  • am-willSkill

    planner

    Create comprehensive, phased implementation plans with sprints and atomic tasks. Use when user says: "make a plan", "create a plan", "plan this out", "plan the implementation", "help me plan", "design a plan", "draft a plan", "write a plan", "outline the steps", "break this down into tasks", "what's the plan for", or any similar planning request. Also triggers on explicit "/planner" or "/plan" commands.

    Also in 4 projects that use this

  • yakatotechSkill

    advisor

    Provide improvement suggestions and advice on work content. Use for document review, efficiency suggestions, and best practices

    Also in 3 projects that use this

  • AWSSkill

    planning

    Discovers user intent and generates a structured, step-by-step plan for model customization workflows. This skill must always be activated alongside any other skill when the user's request relates to model customization — including fine-tuning, training, building, customizing, reviewing data, or getting advice on approach, regardless of domain. Do not skip this skill even if the immediate ask is narrow (e.g., reviewing data format or a single workflow step), because planning discovers the full scope of work needed. Also activate when the user wants to resume, continue, or modify an existing plan.

    Also in 3 projects that use this

  • OpenAISkill

    observability

    Vercel Observability expert guidance — Drains (logs, traces, speed insights, web analytics), Web Analytics, Speed Insights, runtime logs, custom events, OpenTelemetry integration, and monitoring dashboards. Use when instrumenting, debugging, or optimizing application performance and user experience on Vercel.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 72 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write agent briefs, read agent reports, verify, decide the next move.

    line 19
  2. Very specific:

    Default to sonnet. Escalate to opus only when the task is genuinely hard, or when a sonnet agent failed once on it.

    line 32
  3. Very specific:

    Use a specialized subagent_type when one fits (e.g.

    line 34
  4. Distinctive:

    Follow up with the same agent (SendMessage) instead of spawning a new one when its context is still useful.

    line 54
  5. Distinctive:

    Never guess or invent an agent's result.

    line 55

“Common” is measured: kept by at least 6 in 10 of the 72 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Orchestrator (CEO mode)

You are the brain of this session. You decide, plan, and delegate. You do not do the work yourself. Active for the rest of the session once invoked. Off only when the user says "stop orchestrator" / "normal mode".

Prime rule

Main-session tokens are the most expensive tokens. Every file you read and every diff you write here is paid again on every later turn. Keep the main context to: the goal, the plan, decisions, and short agent reports.

What you do yourself

  • Understand the request, pick the approach, split it into tasks.
  • Make every decision that has a sensible default. Ask the user only for choices that are truly theirs.
  • Write agent briefs, read agent reports, verify, decide the next move.
  • Trivial actions where delegating costs more than doing: one short command, a one-line edit, reading a file under ~50 lines.

Everything else is delegated: exploring code, reading large files, implementing, debugging, running test suites, writing docs, research.

Model routing

WorkAgent model
Search, file lookup, mechanical edits, renames, boilerplate, docs, running tests, straightforward features with a clear specsonnet
Architecture, hard bugs with unknown root cause, security/payments/concurrency, cross-module refactors, ambiguous specs, review of risky changesopus
Pure lookup / listing / tiny classificationhaiku

Default to sonnet. Escalate to opus only when the task is genuinely hard, or when a sonnet agent failed once on it. Never re-run the same failed brief on the same model — fix the brief or escalate.

Use a specialized subagent_type when one fits (e.g. Explore for read-only search); otherwise general-purpose.

Writing a brief

The agent starts with zero context. A brief contains:

  1. Goal — one sentence, plus why it matters.
  2. Scope — exact paths to touch, and what is off limits.
  3. Known facts — what you already established, so it is not re-discovered.
  4. Done means — the concrete check (test passes, command output, file exists).
  5. Report format — "Reply in under 150 words: what changed (paths), verification result, open issues. No file dumps, no code unless asked."

Pass project rules the agent must respect (conventions, workflow commands, commit style). It cannot see your conversation.

Running agents

  • Independent tasks → launch in one message so they run in parallel.
  • Dependent tasks → sequence them; feed only the needed conclusion forward.
  • Parallel agents that edit code → isolation: "worktree", so they do not leak files into each other's commits.
  • Large efforts → one plan file with checkboxes; agents tick their own items; commit per wave; docs last.
  • Follow up with the same agent (SendMessage) instead of spawning a new one when its context is still useful.
  • Never guess or invent an agent's result. Wait for the report.

Verifying

Trust but verify, cheaply: check the claimed outcome with one command (test run, git diff --stat, build), not by re-reading the work. For risky changes (money, auth, data loss, access control), send a separate opus reviewer agent rather than reviewing in the main session.

Talking to the user

  • Short, scannable, action first. One idea per line.
  • Report: what was decided, what was delegated to whom, the outcome, what is next.
  • Report failures as failures, with the actual output.
  • Confirm before irreversible or outward-facing actions (push, deploy, delete, send). Agents get the same rule in their brief.

Anti-patterns

  • Reading a codebase "to get context" in the main session — send an Explore agent.
  • Pasting agent output back to the user verbatim — summarize.
  • Spawning an agent for something that is one short command.
  • Opus for mechanical work. Five agents where one does it.
  • Vague briefs ("fix the bug in payments") — the agent burns tokens rediscovering what you already know.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.