Multi agent orchestration

Skill MarchRory/liushi_agent_settings/packages/harness/src/.agents/skills/multi-agent-orchestration

Use when a task may need specialist agents, parallel investigation, role routing, debate, review loops, approval gates, or explicit strategy selection before delegation.From its SKILL.md

Install

npx -y skills add MarchRory/liushi_agent_settings --skill multi-agent-orchestration

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.9 KB, 562 tokens by cl100k_base, as published. Nobody here has run it

Multi-Agent Orchestration

Core Rule

Lead owns strategy selection, specialist startup, conflict resolution, validation, and final claims. Scripts only generate read-only activation packets; they do not start agents or mutate repositories.

Lead Judgment Prompt

Before running any script, analyze the task as Lead:

What outcome is needed, and what would make the work actually correct?
Can Lead finish it alone faster and safer than coordinating specialists?
Which parts are independent, sequential, uncertain, risky, or validation-heavy?
What evidence is already available, and what evidence must be collected before delegation?
Which strategy fits the task shape, and why are the nearest alternatives worse?
What must specialists not do, including permissions, scope, files, tools, and stop conditions?

Use lead_analysis to capture the task decomposition and lead_strategy_reason to capture the chosen strategy rationale. Structured context fields are useful when they make checks repeatable, but they do not replace Lead judgment. If the script recommendation conflicts with the Lead analysis, do not blindly follow the script; record the override reason and use the safer or more evidence-grounded path.

Workflow

Start lead-only unless the task signals real coordination value.
Write a compact task context with task_summary, lead_analysis, lead_strategy_reason, observed signals, risks, and validation path.
Read references/activation-rules.md when strategy choice is not obvious.
Use references/strategy-registry.toml as the machine-readable strategy pool.
Run npm run select -- --context-file context.json or a strategy script with task-context JSON.
Review the packet: if evidence is missing, over-delegated, unsafe, or less accurate than Lead judgment, revise context or override with a recorded reason.
Choose models in real time, then start specialists manually with scoped handoffs.
Use references/sop.md, references/strategy-playbooks.md, and references/anti-patterns.md before broad delegation.

Resource Map

references/strategy-registry.toml: strategy definitions and required evidence.
references/model-policy.toml: model selection hints, not hard-coded assignments.
references/output-schema.md: activation packet schema.
references/platform-compatibility.md: Windows/macOS command and runtime notes.
references/research-basis.md: industry, academic, and open-source basis for each strategy.
references/strategy-playbooks.md: explicit context fields and SOP for each strategy.
scripts/: read-only selector and per-strategy packet generators.
fixtures/, schemas/, and snapshots/: deterministic validation spine for strategy activation behavior.