Discovery, Routing and Delegation

Match need to advertised skill

A client agent needs a way to find candidates: a configured list of trusted agent URLs, an internal registry or catalogue of Agent Cards, or direct knowledge between partners. Routing then compares the job with the cards' skills and descriptions, ideally using exact ids or tags first and a model only for ambiguous cases. Delegation decisions should consider trust (is this agent approved for this data?), cost and latency, and fallbacks (what if it is down or declines?). If no agent advertises the needed skill, say so and escalate to a human rather than guessing. Pass the remote agent only the data it needs, since it operates outside your boundary.

Skill-based routing, run

I ran this with plain Python 3 (standard library only). Refund work maps to the billing agent and compliance work to the legal agent. A request for "poetry" matches no advertised skill, so the code escalates to a human instead of guessing.

AGENTS = {
    "billing": {"skills": {"invoice", "refund"}},
    "legal":   {"skills": {"contract", "compliance"}},
    "support": {"skills": {"faq", "troubleshoot"}},
}

def discover(needed_skill):
    return [name for name, card in AGENTS.items() if needed_skill in card["skills"]]

for skill in ("refund", "compliance", "poetry"):
    found = discover(skill)
    print(skill, "->", found if found else "no agent advertises this skill; escalate to a human")

Output:

refund -> ['billing']
compliance -> ['legal']
poetry -> no agent advertises this skill; escalate to a human

Quick check: What should happen when no agent advertises the needed skill?

• Send the data to everyone
• Pick a random agent
• Say so and escalate rather than guess
• Ignore the request