feat: add agent identity + onboarding protocol example

[vystartasv]

What This Adds

A working example at atomic-examples/agent-identity/ that demonstrates cryptographic agent identity and onboarding — the missing layer for making atomic agents interoperable beyond a single process.

The Two Protocols

Identity Protocol (L2)

Each agent gets a permanent, verifiable identity — an Ed25519-style keypair. The public key is the agent's identifier; the private key signs handoff tokens and capability manifests.

agent = AgentIdentity.create("research-agent-01")
print(agent.fingerprint())  # 1971302321625375

Onboarding Protocol (L1)

Before an agent is trusted with work, an authority issues a signed certificate listing verified capabilities. This is the "prove you can do the job before I give you the job" step.

cert = OnboardingCertificate(
    agent_id=agent.agent_id,
    authority=authority.agent_id,
    capabilities=["web_search", "document_analysis"],
)
cert.signature = authority.sign_claim(cert)
assert cert.is_valid(authority)  # ✅

Why This Matters for Atomic Agents

Atomic agents ("one agent, one job") are composable by design. But composition without identity is fragile:

• ❌ Agent A hands off to "some agent" → no audit trail, no verification
• ✅ Agent A signs a handoff to Agent B (public key verified) → cryptographic proof of the chain

Identity turns a collection of agents into a verifiable pipeline.

What's in This PR

• main.py — 150-line demo: identity creation, onboarding, signed handoff, verification
• pyproject.toml / __init__.py — standard example packaging (matches existing pattern)
• README.md — explains the concept, links to the full protocol spec

Zero new dependencies. The demo uses stdlib only. The full production SDK (pip install works-with-agents) adds Ed25519 via pynacl for real cryptographic signing — but the concept and protocol structure are the same.

Run It

cd atomic-examples/agent-identity
python agent_identity/main.py

References

• Identity Protocol spec: https://workswithagents.com/specs/identity.md (CC BY 4.0)
• Original proposal: Proposal: Agent Identity + Onboarding protocols for interoperable atomic agents #251
• Related bug fix: fix: prevent None in MCP prompt text extraction crashing str.join #252 (same session, separate concern)