agntcy/shadi/assembly

ASSEMBLY — model the problem together

You are one peer on a SHADI mesh. MCP is off. This phase is group modeling, not a single agent solving.

Understand the problem with your peers. Propose a system model: who the agents are, what they exchange, and what “better” means. Then derive the update rule: how one agent computes its next value from what it can see.

You may name a class that is not one of the three paper examples (Preference Aggregation, Supply Chain Cascades, Sustainable Resource Allocation). Those three are examples, not a closed set.

The update rule

In CONVERGE you apply the rule you give here yourself, every epoch. Nothing else computes your next value, and a wrong rule is applied as given.

Write it over your class's named local quantities only:

Use numbers, + - * /, parentheses, max, min, abs, and clamp(x, low, high). For example max(0, a - b) / (1 + c), with your class's names in place of a, b and c.

Deriving the rule is not computing the answer. Give the rule, never a solution.

Line protocol

End with exactly two lines:

CLASS <name> UPDATE <expression>

<name> is preference, cascade, resource, or another short token if you believe the problem is a different class. Then NEXT goose-<n> or DONE. No other text.

Forbidden

• Do not compute or mention a global target vector, z*, a full demand

path, bullwhip, or a terminal stock forecast.

• Do not start MCP, desktop, or Summon.
• Do not invent agentbridge or Goose flags.

Install this skill

Copy the skills/assembly/ folder to the host skills directory. The folder name must be assembly.