agent-orchestration

agent-orchestration is a skill for Claude Code from CassetteTapeCrackle/claude-starters. It costs 59 tokens per session (647 once invoked), scanned A, original, MIT.

A policy for deciding whether a coding task should be done directly or given to a specialist agent. A specialist agent is a separate helper focused on a particular kind of work.

In plain words
What is it for?
It is for choosing between inline work, delegation, or no delegation based on the task and the agents available in the project.
Why use it?
It reduces unnecessary delegation while identifying tasks where a specialist could improve planning, debugging, design, or review.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: reads .claude/ paths; mentions CLAUDE.md; mentions subagents.

Part of the claude-starters plugin — 9 skills, 1 command, 8 agents, 2 hooks shipped together

Good fit It is for choosing between inline work, delegation, or no delegation based on the task and the agents available in the project.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/cassettetapecrackle/claude-starters/agent-orchestration
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add CassetteTapeCrackle/claude-starters --skill agent-orchestration
Clone the repo
git clone --depth 1 https://github.com/CassetteTapeCrackle/claude-starters

Made for: Claude Code.

Or install claude-starters, the plugin that ships this one along with the rest of its 9 skills, 1 command, 8 agents, 2 hooks.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for agent-orchestration

README.md
[![agentmods](https://agentmods.dev/badge/skills/cassettetapecrackle/claude-starters/agent-orchestration/github.svg)](https://agentmods.dev/skills/cassettetapecrackle/claude-starters/agent-orchestration)
Your own site
<a href="https://agentmods.dev/skills/cassettetapecrackle/claude-starters/agent-orchestration"><img src="https://agentmods.dev/badge/skills/cassettetapecrackle/claude-starters/agent-orchestration/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for agent-orchestration

Your own site · 80×15
<a href="https://agentmods.dev/skills/cassettetapecrackle/claude-starters/agent-orchestration"><img src="https://agentmods.dev/badge/skills/cassettetapecrackle/claude-starters/agent-orchestration.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 59 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 647 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00059 $0.00647
Opus 5 $0.00030 $0.00324
Sonnet 5 $0.00012 $0.00129
Haiku 4.5 $0.00006 $0.00065

Measured 11d ago against content hash a0ab448ded31, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

agent-orchestration scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/agent-orchestration/SKILL.md · 35 lines

How it starts

The opening of the file, as written. The whole thing — 35 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agent orchestration (the dispatcher policy)

You are the orchestrator. You choose between doing work yourself and delegating to a specialist agent. Optimize for fewer tokens and better output — most of the time that means doing it yourself or skipping.

Candidates come from two channels

  • Reactive — the Stop hook surfaced a code-risk candidate (e.g. unsafe-auditor, rt-safety-auditor, any-eliminator) because changed code matched a risk pattern.
  • Proactive — the phase of the current task maps to a specialist:
    • ideate → plugin-ideator / idea-brainstormer / competitor-analyst
    • spec/plan → prd-writer / spec-critic
    • research an unknown → deep-researcher
    • design UI → ui-designer / ux-critic / plugin-ui-designer / design-system-author
    • debug → debugger / regression-hunter
    • audit quality → the per-stack auditors + starter-conformance-checker
    • ship → technical-writer / release-copywriter / demo-script-writer / naming-agent

Only consider agents that actually exist for this project: the ones in .claude/agents/ (scoped) plus the global ones. Don't invent agents.

The decision — skip / inline / delegate

For each candidate, pick the cheapest option that gets the job done well:

  • SKIP if the agent's concern isn't really implicated, or the change is trivial (docs, a rename, a one-line tweak). This is the most common answer. Don't dispatch to prove diligence.
  • INLINE (do it yourself) if the surface is small and local (a few lines / one file) and you already have the context. Spinning a subagent would cost more than just doing it.
  • DELEGATE only when it clearly pays: a broad scan (many files / a whole module), work that would pull a lot into your context (isolation is the payoff — the subagent's tokens stay out of your window), or a well-scoped specialist job whose prompt does better than you would ad hoc.

Rules

  • Bias to skip/inline. When unsure and the surface is small, inline.
  • One agent at a time; don't chain speculative delegations. Delegate, read the result, then decide the next step.
  • Respect the project's own rules (its CLAUDE.md) — e.g. don't let an agent touch DSP on a non-DSP task.
  • After you've handled a surfaced candidate, don't re-evaluate the same unchanged diff.

Read the full file on GitHub · 35 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 11d ago First seen · 35 lines · 59 tokens per session scan A a0ab448ded31

Subscribe to this mod's changes

agent-orchestration is a skill published in the GitHub repository CassetteTapeCrackle/claude-starters (1 stars, last pushed 22d ago), licensed MIT. It adds 59 tokens to every session and 647 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

open-problem

Use when the user points at a hard, unsolved, or open problem in any field — mathematics, physics, biology, ML, engineering — and wants a real attack on it. Runs a sustained multi-agent assault: establishes the state of play from primary sources, hunts for the two literatures nobody has combined, generates and kills…

infiniV/ultra-ml-intern · 141 tokens

grounding

Use before writing, reviewing, or debugging any code that uses a specific ML model (DINOv3, SAM 2, Whisper, Qwen3-Embedding, SigLIP 2…), whenever a model-provenance archive for it exists locally. Loads that archive's real source — checkpoint ids, API signatures, preprocessing constants, training recipe — so the code…

infiniV/ultra-ml-intern · 147 tokens

engineering-discipline

Behavioral guidelines to reduce common agent coding mistakes — thinking before coding, simplicity, surgical changes, goal-driven execution, and mandatory verification for data/BI work. Adapted from andrej-karpathy-skills (MIT), with a data/BI engineering layer added. Use this skill whenever writing, editing, or…

mukilankarthik/self-awareness · 86 tokens

self-awareness

Behavioral calibration system that helps Claude catch its own failure patterns in real-time. Triggers on EVERY response to run self-checks for verbosity compensation, hallucination risk, missed frustration signals, sycophantic patterns, and constraint violations. Use this skill continuously—it modifies how Claude…

mukilankarthik/self-awareness · 67 tokens

fable-method

Use when a task carries real uncertainty — multi-step builds, debugging where the first theory may be wrong, research or analysis with claims to verify, anything touching data/APIs/files not yet opened, or work that will be handed off. Also when work keeps failing or stalling, before declaring anything done, or when…

debabsah/fable-method · 111 tokens

model-provenance

Use when the user names a specific ML model (e.g. DINOv3, SAM 2, Whisper, Qwen2-VL) and wants its real/official code, training recipe, or papers found, verified, or archived locally for grounded coding. Triggers include "find the real code for this model", "harvest DINOv3", "store the model's code and papers locally"…

infiniV/ultra-ml-intern · 134 tokens