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.
npx skills add sharpdeveye/maestro --skill composegit clone --depth 1 https://github.com/sharpdeveye/maestroWrote 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.
[](https://agentmods.dev/skills/sharpdeveye/maestro/compose)<a href="https://agentmods.dev/skills/sharpdeveye/maestro/compose"><img src="https://agentmods.dev/badge/skills/sharpdeveye/maestro/compose/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.
<a href="https://agentmods.dev/skills/sharpdeveye/maestro/compose"><img src="https://agentmods.dev/badge/skills/sharpdeveye/maestro/compose.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00020 | $0.00661 |
| Opus 5.5 | $0.00008 | $0.00264 |
| Sonnet 5.5 | $0.00004 | $0.00132 |
| Haiku 4.5 | $0.00002 | $0.00066 |
Grade A, and why
compose 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 1mo 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.
How it starts
The opening of the file, as written. The whole thing — 87 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MANDATORY PREPARATION
Invoke /agent-workflow — it contains workflow principles, anti-patterns, and the Context Gathering Protocol. Follow the protocol before proceeding — if no workflow context exists yet, you MUST run /teach-maestro first. Consult the agent-architecture reference in the agent-workflow skill for topology patterns and when multi-agent is justified.
Design a multi-agent system. But first — are you sure you need one?
Step 1: Pre-Composition Check
Answer these before proceeding:
- Has a single agent been tried and failed? (If no, try single agent first)
- What specific limitation requires multiple agents? (If you can't name it, you don't need multi-agent)
- Is the cost/latency increase justified? (Multi-agent = 2-10x cost and latency)
If you can't articulate a specific limitation, use /amplify on the single agent instead.
Step 2: Design the Topology
Choose the right architecture pattern (consult the agent-architecture reference in the agent-workflow skill):
For each agent in the system, define:
## Agent: [Name]
Role: [One sentence]
Responsibilities: [What it does]
Boundaries: [What it does NOT do]
Tools: [List of tools this agent has access to]
Input: [What it receives]
Output: [What it produces]
Step 3: Design Handoffs
For each agent-to-agent connection:
## Handoff: [Agent A] → [Agent B]
Trigger: [When does A hand off to B?]
Payload: [What data is passed?]
Expected response: [What does A expect back?]
Timeout: [How long to wait?]
Failure handling: [What if B fails?]
Step 4: Design the Supervisor
Every multi-agent system needs a supervisor:
- Monitors agent health and performance
- Routes tasks to appropriate agents
- Handles failures and escalation
- Enforces global constraints (budget, time, quality)
Composition Checklist
- Each agent has a clear, non-overlapping role
- Handoff protocols are defined for every connection
- A supervisor pattern is in place
- Cost/latency budget accounts for all agents
- Failure modes are handled at every handoff point
- The system can be understood by reading the topology diagram
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.
- 1mo ago First seen · 87 lines · 20 tokens per session scan A f3b306b99c39
compose is a skill published in the GitHub repository sharpdeveye/maestro (591 stars, last pushed 5mo ago), licensed MIT. It adds 20 tokens to every session and 661 once invoked, about $0.0001 per session on Opus 5.5. 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 2026-08-30.
Other skills, from other repositories
aider-delegate
Delegate coding tasks to Aider (aider) only when the user explicitly requests it, while the orchestrator retains review and landing responsibility.
agy-delegate
Delegate coding tasks to the Google Antigravity CLI (agy) only when the user explicitly requests it, while the orchestrator retains review and landing responsibility.
agent-self-scheduling
Schedule AI agent runs with cron, loops, or external clocks while avoiding unsafe tight autonomous timers.
deep-interview
Socratic deep interview with mathematical ambiguity gating before execution.
plan
Strategic planning with optional interview workflow.
web-clone
URL-driven website cloning with visual + functional verification.